Certified Cloud Native Applied

Generative AI Engineer
Master the Future

Build Custom GPTs, AI Agents, Humanoids, and Fine-Tune LLMs

Version: 12.5 (Implementation and adoption starting from August 1, 2024)

Today's pivotal technological trends are Cloud Native (CN), Generative AI (GenAI),
and Physical AI. Cloud Native technology offers a scalable and dependable platform
for application operation, while AI equips these applications with intelligent, human-
like capabilities. Physical AI aims to bridge the gap between digital intelligence and
physical capability, creating systems that can understand and interact with the world
in a human-like manner. Our aim is to train you to excel as a Cloud Native Applied
Generative and Physical AI developer globally.

The Cloud Native Applied Generative AI Certification program equips you to create
leading-edge Cloud Native AI and Physical AI solutions using a comprehensive
cloud-native, AI, and Physical AI platform.

Everything will soon be represented by a conversational interface, or to put it another

way, a personal AI, we will cover it extensively in this program. () Currently, OpenAI
Custom GPT Platform is the best platform to develop personal AI.
We will also be covering AI agents, which are autonomous programs or entities that
perceive their environment through sensors, process this information, and take
actions to achieve specific goals or tasks. They can operate independently, adapt to
changing conditions, and make decisions based on their observations and
objectives.

Material to Understand the Coming AI Age:

● Watch the Overview Video of Our Program
● Watch AGI could Double GDP
● Watch Personal AI Short Video
● The Future Is Agentic
● The INSANE Race for AI Humanoid Robots
● What Is an AI Anyway? Mustafa Suleyman
● The Coming Wave: Technology, Power, and the 21st Century’s Greatest
Dilemma
● The Worlds I See: Curiosity, Exploration, and Discovery at the Dawn of AI
● Ethan Mollick’s Substack
● David Autor Lecture
● Conversation between Suleyman, Yuval Noah Harari, and Zanny Minton
Beddoes

This twenty one month program equips you with the skills to thrive in the age of
Generative AI (GenAI), Physical AI, and cloud native computing (CN). You will
become an expert Custom GPT, AI Agent, and Humanoid Robotics Developer. The
program is divided into two levels: foundation level and professional level. Students
will be able to start working after completing the foundation level. They will
continue their professional level studies while working.

Why This Program?

● Cutting-Edge Skills: Develop in-demand skills to build intelligent, scalable

cloud applications using Generative AI and Cloud Native technologies.
● Industry-Ready: Prepare for global certifications, startup and freelance
opportunities after just six months.
● Future-Proof Your Career: Stay ahead of the curve in a rapidly evolving tech
landscape.

What You'll Learn:

● Custom GPTs and Multi AI Agent Systems: Learn to fine-tuning

foundational AI models, and market them in GPT stores. Learn key principles
of designing effective AI agents, and organising a team of AI agents to
perform complex, multi-step tasks. Apply these concepts to automate
common business processes.

2
 ● Develop AI Powered Microservices: Master Python, build APIs using
FastAPI, SQLModel, Postgres, Kafka, Kong, and leverage cutting-edge GenAI
APIs like OpenAI, and Open Source AI LLMs.
● Cloud Native Expertise: Design and deploy cloud-native applications using
Docker, DevContainers, TestContainers, Kubernetes, Terraform, and GitHub
Actions.
● Distributed System Design: Designing systems that run on multiple
computers (or nodes) simultaneously, interacting and coordinating their
actions by passing messages over a network.
● Designing AI Solutions using Design Thinking and Behaviour Driven
Development (BDD): We will learn to leverage these methodologies to create
AI solutions that are not only technically sound but also highly user-centric
and aligned with real-world needs.
● Fine-Tuning Open-Source Large Language Models using PyTorch, and
Fast AI: We will learn to fine-tuning of open-source Large Language Models
(LLMs) like Meta LLaMA 3 using PyTorch and Fast AI, with a focus on cloud-
native training and deployment. We will set up development environments,
preprocess data, fine-tune models, and deploy them using cloud native
platforms.
● Physical AI and Humanoid Robotics: We will learn to design, simulate, and
deploy advanced humanoid robots capable of natural interactions.

Flexible Learning:

● Earn While You Learn: Start freelancing or contributing to projects after the
third quarter.

Program Structure (Foundation: 3 + Professional: 4 = Total: 7 Quarters):

Foundation Level (3 Quarters)

● Quarter 1: Fundamentals of Prompt Engineering, Docker, GitHub, and
Modern Python Programming

We begin the course by understanding the basics of GenAI and Prompt

Engineering. Then we will understand the basics of Linux, Docker, VSCode,
Devcontainer, and GitHub. The main focus will be on mastering the
fundamentals of Modern Python with Typing, the go-to language for AI.

○ Certification:
■ Certified Professional Python Programmer (CPPP1)

Learning Repo: https://github.com/panaversity/learn-cloud-native-modern-

python

● Quarter 2: Applied Generative AI Fundamentals: Prompt Engineering,

Developing Custom GPTs and Multi AI Agent Systems

3
 With this course, you’ll start by building a strong understanding of generative
AI and learn how to apply Large language models (LLMs) and diffusion
models practically. We will introduce a set of principles known as prompt
engineering, which will help developers to work efficiently with AI. Learn to
create custom AI models and GPTs using OpenAI, Azure, and Google
technologies. Use open source libraries, like Langchain, CrewAI, and
LangGraph to automate repeatable, multi-step tasks and automate business
processes that are typically done by a group of people.

Certifications:

■ Microsoft Certified: Azure AI Engineer Associate

■ Certified crewAI Engineer

Learning Repo: https://github.com/panaversity/learn-applied-generative-ai-

fundamentals/

● Quarter 3: Cloud Native AI Powered Microservices Design, Development,

and Deployment:

Build scalable AI Powered APIs using FastAPI, Postgres, Kafka, Kong, GenAI
APIs like OpenAI Chat Completion APIs, Assistant APIs, LangChain and
Open Source AI LLMs, develop them using Containers and Dev Containers,
and deploy them using Docker Compose locally and Kubernetes Powered
Serverless Container Services on the cloud.

We will also learn to integrate design thinking and Behavior-Driven

Development (BDD) in developing AI systems. We will learn to create AI
solutions that are deeply aligned with user needs and expectations. Design
thinking ensures a thorough understanding of the user and problem space,
while BDD provides a structured approach to defining and validating the
desired behaviours of the AI system. Together, these methodologies lead to
the development of AI solutions that are not only technically robust but also
highly user-centric and effective in solving real-world problems.

○ Certifications:
■ PostgreSQL 13 Associate Certification
■ Confluent Certified Developer for Apache Kafka (CCDAK)
■ Design Thinking Professional Certificate (DTPC)
■ Test and Behavior Driven Development (TDD/BDD)

Learning Repo:

https://github.com/panaversity/learn-cloud-native-ai-powered-microservices/

We Will Be Using Microsoft Azure as our Default Cloud Platform

4
Amazon is still the cloud king based on market share. But many analysts
agree: In the battle for the cloud, AI is now a game-changer — and Amazon's
main competitors, particularly Microsoft, have the momentum.

In our program we will be using Azure as our default provider for teaching and
deployment. We will be using using these services:

Get a free Azure Account now:

https://azure.microsoft.com/en-us/free
Note: Use GitHub Account to start an Azure free trial

Azure Container Apps (We will Start from this service using Dapr and Keda)
https://azure.microsoft.com/en-us/products/container-apps
Get started with the free tier: The first 180,000 vCPU per second, 360,000
GiB/s, and 2 million requests each month are free.
Watch: https://www.youtube.com/watch?v=0HwQfsa03K8
Deploy: https://learn.microsoft.com/en-us/azure/container-apps/code-to-cloud-
options

Azure Container Registry

https://azure.microsoft.com/en-us/products/container-registry/

Deploy to Azure Container Apps with GitHub Actions

https://learn.microsoft.com/en-us/azure/container-apps/github-actions

Azure Kubernetes Service (AKS)

https://azure.microsoft.com/en-us/products/kubernetes-service

GitHub
https://azure.microsoft.com/en-us/products/github/

GitHub Actions for AKS

https://learn.microsoft.com/en-us/azure/aks/kubernetes-action

Azure OpenAI Service

https://azure.microsoft.com/en-us/products/ai-services/openai-service

Azure Database for PostgreSQL

https://azure.microsoft.com/en-us/products/postgresql/

Kafka
https://cloudatlas.me/5-different-ways-you-can-run-apache-kafka-on-azure-
973a18925ac7

5
Professional Level (4 Quarters)

● Quarter 4: Generative AI with PyTorch:

Generative AI tools like ChatGPT, Gemini, and DALL-E have revolutionised

our professional landscape. This hands-on course, “Master Generative AI with
PyTorch,” guides you through the exciting process of building and training AI
models using Python and the versatile, open-source PyTorch framework, all
with the hardware you already have. You’ll delve into the core concepts of
Generative Adversarial Networks (GANs), Transformers, Large Language
Models (LLMs), variational autoencoders, diffusion models, LangChain, and
more. Along the way, you’ll gain practical experience and a deep
understanding of these cutting-edge technologies.

Learning Repo: https://github.com/panaversity/genai-with-pytorch

● Quarter 5: Fine-Tuning Open-Source Large Language Models:

This comprehensive course is designed to guide learners through the process

of fine-tuning open-source Large Language Models (LLMs) such as Meta
LLaMA 3 using PyTorch, with a particular emphasis on cloud-native training
and deployment. The course covers everything from the fundamentals to
advanced concepts, ensuring students acquire both theoretical knowledge
and practical skills.

The journey begins with an introduction to LLMs, focusing on their

architecture, capabilities, and the specific features of Meta LLaMA 3. Next, the
course dives into PyTorch fundamentals, teaching students how to perform
basic operations with tensors and build simple neural networks. This
foundation is crucial for understanding the mechanics behind LLMs. Data
preparation is a crucial aspect of training models. The course covers
comprehensive data collection and preprocessing techniques, such as
tokenization and text normalisation. These steps are essential for preparing
datasets suitable for fine-tuning LLMs like Meta LLaMA 3. Through practical
exercises, students learn how to handle and preprocess various types of text
data, ensuring they can prepare their datasets for optimal model performance.

Fine-tuning Meta LLaMA 3 with PyTorch forms a significant part of the course.
Students will delve into the architecture of Meta LLaMA 3, learn how to load
pre-trained models, and apply fine-tuning techniques. The course covers
advanced topics such as regularisation and optimization strategies to
enhance model performance. Practical sessions guide students through the
entire fine-tuning process on custom datasets, emphasising best practices
and troubleshooting techniques.

6
 A critical aspect of this course is its focus on cloud-native training and
deployment using Nvidia NIM. Furthermore, students learn how to deploy
models using Docker and Kubernetes, set up monitoring and maintenance
tools, and ensure their models are scalable and efficient.

To round off the learning experience, the course includes an in-depth

segment on exporting models for inference and building robust inference
pipelines. Students will deploy models on cloud platforms, focusing on
practical aspects of setting up monitoring tools to maintain model performance
and reliability.

The course culminates in a capstone project, where students apply all the
skills they have learned to fine-tune and deploy Meta LLaMA 3 on a chosen
platform. This project allows students to demonstrate their understanding and
proficiency in the entire process, from data preparation to cloud-native
deployment.

Learning Repo:

https://github.com/panaversity/learn-fine-tuning-llms

● Quarter 6: Physical AI and Humanoid Robotics Development:

Artificial intelligence (AI) has experienced remarkable advancements in recent

years. However, the future of AI extends beyond the digital space into the
physical world, driven by robotics. This new frontier, known as “Physical AI,”
involves AI systems that can function in the real world and comprehend
physical laws. This marks a notable transition from AI models confined to
digital environments. Humanoid robots are poised to excel in our human-
centred world because they share our physical form and can be trained with
abundant data from interacting in human environments.

This course provides an in-depth exploration of humanoid robotics, focusing

on the integration of ROS 2 (Robot Operating System), Open Source Meta
Llama 3, and OpenAI technologies. Students will learn to design, simulate,
and deploy advanced humanoid robots capable of natural interactions. The
curriculum covers essential topics such as ROS 2 for robotic control,
simulations with Gazebo and Unity, and using OpenAI’s GPT models for
conversational AI. Through practical projects and real-world applications,
students will develop the skills needed to drive innovation in humanoid
robotics.

Learning Repo: https://github.com/panaversity/learn-physical-ai-humanoid-

robotics

● Quarter 7: Kubernetes and Distributed System Design:

7
 Master Kubernetes, Terraform, and GitHub Actions to deploy your AI APIs
and microservices in the cloud. We will cover distributed system design
involving creating systems that are distributed across multiple nodes, focusing
on scalability, fault tolerance, consistency, availability, and partition tolerance.

Certifications:

■ Certified Kubernetes Application Developer (CKAD)

■ HashiCorp Certified: Terraform Associate

Learning Repo: https://github.com/panaversity/learn-kubernetes

Frontend Specialisation

● Quarter 8: Front-end Web GUI Development using Next.js and

TypeScript (Optional):

Next.js is designed to handle complex front-end applications well, making it a

good fit for AI applications that might grow in features and data usage over
time. Next.js offers features like API routes and file-based routing, which can
streamline development for AI applications that need to interact with backend
APIs and manage different application views. While Next.js and TypeScript
aren't the only options for building AI application frontends, their focus on
performance, scalability, and developer experience makes them a compelling
choice for many developers.

Learning Repo:
https://github.com/panaverse/learn-nextjs

8
In 2015, Klaus Schwab, founder of the World Economic Forum, asserted that we
were on the brink of a “Fourth Industrial Revolution,” one powered by a fusion of
technologies, such as advanced robotics, artificial intelligence, and the Internet of
Things.

“[This revolution] will fundamentally alter the way we live, work, and relate to one
another,” wrote Schwab in an essay published in Foreign Affairs. “In its scale, scope,
and complexity, the transformation will be unlike anything humankind has
experienced before.”

Generative AI is set to revolutionise our daily lives and work environments.

According to McKinsey & Company, generative AI could contribute an annual
economic value of $2.6 trillion to $4.4 trillion across various sectors by enhancing
automation, bolstering decision-making, and providing personalised experiences.

Investor Cathie Wood predicts that the market for humanoid robots could grow to $1
trillion by 2030.

Cloud native is an approach in software development that enables application

creation, deployment, and management in cloud environments. It involves
constructing applications as a collection of small, interconnected services known as
microservices, a shift from traditional monolithic structures. This modular approach

9
enhances the agility of cloud-native applications, allowing them to operate more
efficiently with fewer resources.

Cloud Native has already been adopted by the majority of the companies, by 2024,
more than 90% of global organisations will be running containerized applications in
production. The adoption of Docker and Kubernetes has seen significant growth over
recent years. As of 2022, about 61% of organisations reported using Kubernetes for
container orchestration. This number has been steadily increasing as more
companies realise the benefits of these technologies for managing containerized
applications  .

Technologies such as Kubernetes, Docker, serverless containers, APIs, SQL

Databases, and Kafka support developers in swiftly constructing cloud-native
applications. These tools offer a standardised platform for application development
and management across various cloud services like Azure, Google Cloud, and AWS.

This revolution is pivotal for technology and job landscapes, making it essential
knowledge in fast-evolving tech cycles. The rapid emergence of Gen AI-powered
and Physical AI technologies, and the evolving demand for skills necessitate
extensive and timely professional training.

Vertical Specialization Level

Students will have the option of selecting one of the following specialisations after
the completion of sixth quarter i.e. in the seventh quarter:

1. Healthcare and Medical GenAI Specialization

2. Web3, Blockchain, and GenAI Integration Specialization
3. Metaverse, 3D, and GenAI Integration Specialization
4. GenAI for Accounting, Finance, and Banking Specialization
5. GenAI for Engineers Specialization
6. GenAI for Sales and Marketing Specialization
7. GenAI for Automation and Internet of Things (IoT) Specialisation
8. GenAI for Cyber Security

Common Questions (FAQs) with Detailed Answers

1. What is Cloud Native Applied Generative AI Engineering?

Cloud Applied Generative AI Engineering (GenEng) is the application of
generative AI technologies to solve real-world problems in the cloud.
● Generative AI is a type of artificial intelligence that can create new data
or content from existing data.

10
 ● Cloud Native computing is the delivery of computing services—
including servers, storage, databases, networking, software, analytics,
and intelligence—over the Internet (“the cloud”).

By combining generative AI with cloud native computing, businesses can

solve a variety of problems, such as:
● Creating personalised experiences for customers
● Automating tasks
● Improving decision-making
● Detecting fraud
● Developing new products and services

The potential applications of cloud native-applied generative AI are endless.

As generative AI and cloud native computing continue to develop, we can
expect to see even more innovative and groundbreaking uses for this
technology.

2. How valuable are the Cloud Native Applied Generative AI developers?

Developers with expertise in Cloud Native Applied Generative AI were in
extremely high demand due to the increasing adoption of GenAI technologies
across various industries. However, the supply of developers skilled
specifically in this niche area might not have been as abundant compared to
more generalised AI or cloud computing roles.

The demand for AI developers, especially those proficient in applying

generative AI techniques within cloud native environments, has been rising
due to the growing interest in using AI for creative applications, content
generation, image synthesis, natural language processing, and other
innovative purposes.

According to some sources, the average salary for a Cloud Native Applied
Generative AI developer in the global market is around $150,000 per year.
However, this may vary depending on the experience level, industry, location,
and skills of the developer. For example, a senior Cloud Applied Generative
AI developer with more than five years of experience can earn up to $200,000
per year. A Cloud Applied Generative AI developer working in the financial
services industry can earn more than a developer working in the
entertainment industry. A Cloud Applied Generative AI developer working in
New York City can earn more than a developer working in Dubai. In general,
highly skilled AI developers, especially those specialising in applied
generative AI within cloud environments, tend to earn competitive salaries that
are often above the average for software developers or AI engineers due to
the specialised nature of their skills. Moreover, as generative AI technology
becomes more widely adopted and integrated into various products and

11
 services, the demand for Cloud Applied Generative AI developers is likely to
increase.

Therefore, Cloud Applied Generative AI developers are valuable professionals

who have a bright future ahead of them. They can leverage their creativity and
technical skills to create innovative solutions that can benefit various
industries and domains. They can also enjoy very competitive salary and
career growth opportunities.

3. What is the potential for Cloud Applied Generative AI Developers to start

their own companies?
Cloud Applied Generative AI Developers have a significant potential to start
their own companies due to several factors:

1. Emerging Field: Generative AI, particularly when applied within cloud

environments, is still an evolving field with immense potential for innovation.
Developers who understand the intricacies of both generative AI techniques
and cloud technologies can identify unique opportunities to create novel
products, services, or solutions.
2. Market Demand: There is a growing demand for AI-driven applications,
especially those that involve generative capabilities such as image
generation, content creation, style transfer, etc. Developers with expertise in
this area can leverage this demand to create specialized products that cater
to specific industries or consumer needs.
3. Innovation and Differentiation: The ability to develop unique and innovative
solutions using generative AI in the cloud can set apart these developers'
startups from more conventional companies. They can explore new ways of
generating content, enhancing user experiences, or solving complex problems
with AI-generated solutions.
4. Access to Cloud Resources: Cloud platforms provide scalable and cost-
effective resources that are crucial for AI development. Developers starting
their own companies can leverage cloud services to access powerful
computing resources, storage, and AI-related services without significant
upfront investment.
5. Entrepreneurial Opportunities: Developers with entrepreneurial spirit and a
deep understanding of AI technologies can identify gaps in the market and
build startups to fill those gaps. They can create platforms, tools, or services
that simplify the adoption of generative AI for businesses or developers.
6. Collaboration and Partnerships: These developers can collaborate with
other experts in AI, domain specialists, or businesses to create innovative
solutions or explore new application areas for generative AI in the cloud.

However, starting a company, especially in a specialised field like Cloud

Applied Generative AI, requires more than technical expertise. It also

12
 demands business acumen, understanding market needs, networking,
securing funding, managing resources effectively, and navigating legal and
regulatory landscapes.

Successful entrepreneurship in this domain involves a combination of

technical skills, innovation, a deep understanding of market dynamics, and the
ability to transform technical expertise into viable products or services that
address real-world challenges or opportunities.

Developers aspiring to start their own companies in the Cloud Applied

Generative AI space can do so by conducting thorough market research,
networking with industry experts, building a strong team, and developing a
clear business plan that highlights the unique value proposition of their
offerings.

To sum up, the potential for Cloud Applied Generative AI Developers to start
their own companies is high.

● Generative AI is a rapidly growing field with a high demand for skilled

professionals.
● The Certified Generative AI (GenEng) Developer and Engineering
Program provides students with the skills and knowledge they need to
develop and apply cutting-edge generative AI technologies.
● The program also teaches students how to start and run a successful
business.
● Graduates of the program will be well-positioned to start their own
companies and capitalise on the growing demand for generative AI
solutions.

4. Is the program not too long, twenty one months is a long time?
The length of the program is twenty one months which is broken down into
seven quarters of three months each. The program covers a wide range of
topics including Python, GenAI, Microservices, Database, Cloud
Development, Fine-tuning, DevOps, GPTs, AI Agents, and Humanoids. The
program is designed to give students a comprehensive understanding of
generative AI and prepare them for careers in this field. Nothing valuable can
be achieved overnight, there are no shortcuts in life.

5. Why don't we use TypeScript (Node.js) to develop APIs instead of using

Python?
We will not use Typescript in GenAI API development because Python is a
priority with the AI community when working with AI and if any updates come

13
 in libraries they will first come for Python. Python is always a better choice
when dealing with AI and API.

● Python is the de facto standard for AI Development.

● TypeScript is a more modern language that is gaining popularity for
Web Development, but Python is more widely used and has a larger
ecosystem of libraries and frameworks available, especially for AI.
● TypeScript is used for web user interfaces, while Python is used for
APIs.
● In the second quarter, students will learn to develop APIs using Python
instead of TypeScript.
● Python is a more commonly used language for AI and API
development, and it has a larger ecosystem of libraries and
frameworks available for these purposes.
● TypeScript is a more modern language that is becoming increasingly
popular for API development also, but it is still not as widely used as
Python, especially for AI applications and development.

6. What is the difference between OpenAI Completion API, OpenAI

Assistant API, Google Gemini Multi-Modal API, and LangChain?
The difference between OpenAI Completion API, OpenAI Assistant API,
Google Gemini Multi-Modal API, and LangChain is that they are different
ways of using artificial intelligence to generate text, images, audio, and video
based on some input, but they have different features and applications. Here
is a summary of each one:

OpenAI Completion API is the most fundamental OpenAI model that

provides a simple interface that’s extremely flexible and powerful. You give it
a prompt and it returns a text completion, generated according to your
instructions. You can think of it as a very advanced autocomplete where the
language model processes your text prompt and tries to predict what’s most
likely to come next. The Completion API can be used for various tasks such
as writing stories, poems, essays, code, lyrics, etc. It also supports different
models with different levels of power suitable for different tasks.

OpenAI Assistant API is an interface to OpenAI's most capable model (gpt-

4) and their most cost-effective model (gpt-3.5-turbo). It provides a simple way
to take text as input and use a model like gpt-4 or gpt-3.5-turbo to generate an
output. The Assistant API allows you to build AI assistants within your
applications. An Assistant has instructions and can leverage models, tools,
and knowledge to respond to user queries. The Assistant API currently
supports three types of tools: Code Interpreter, Retrieval, and Function
calling.

14
 Google Gemini Multi-Modal API is a new series of foundational models built
and introduced by Google. It is built with a focus on multimodality from the
ground up. This makes the Gemini models powerful against different
combinations of information types including text, images, audio, and video.
Currently, the API supports images and text. Gemini has proven by reaching
state-of-the-art performance on the benchmarks and even beating the
ChatGPT and the GPT4-Vision models in many of the tests.
There are three different Gemini models based on their size, the Gemini Ultra,
Gemini Pro, and Gemini Nano in decreasing order of their size.

LangChain is a platform that allows you to interact with various language

models from different providers such as OpenAI, Google Gemini, Hugging
Face Transformers, etc. You can use LangChain to create applications that
leverage the power of natural language processing without having to deal with
the complexity of APIs or SDKs. LangChain provides a user-friendly interface
that lets you choose the model you want to use, customize the parameters
you want to apply, and see the results in real-time.

7. Why don't we use Flask or Django for API development instead of

FastAPI?
● FastAPI is a newer and more modern framework than Flask or
Django. It is designed to be fast, efficient, and easy to use. FastAPI is
also more scalable than Flask or Django, making it a better choice for
large-scale projects.
● FastAPI is also more feature-rich than Flask or Django. It includes
several built-in features that make it easy to develop APIs, such as
routing, validation, and documentation.
● Overall, FastAPI is a better choice for API development than Flask
or Django. It is faster, more scalable, and more feature-rich.

8. Why do we need to learn Cloud technologies in a Generative AI

program?
Cloud technologies are essential for developing and deploying generative AI
applications because they provide a scalable and reliable platform for hosting
and managing complex workloads.

● Cloud computing offers a vast pool of resources that can be

provisioned on demand, which is ideal for generative AI applications
that can be computationally intensive.

15
 ● Cloud providers offer a wide range of services that can be used to
support generative AI applications, including storage, computing,
networking, and machine learning.
● Cloud services are typically more cost-effective than on-premises
infrastructure, which can be a significant advantage for generative AI
applications that are often used for large-scale projects.

The Certified Generative AI (GenEng) Developer and Engineering Program

teaches you how to use cloud native services, including containers and
Kubernetes, to deploy your applications to the cloud. You will also learn how
to use Docker containers to package and deploy your applications, and how
to use Terraform to manage your cloud infrastructure.

By the end of the program, you will be able to:

● Use Docker containers to package and deploy your applications
● Develop and deploy generative AI applications to the cloud
● Manage your cloud infrastructure using Terraform

9. What is the purpose of Docker Containers and what are the benefits of
deploying them with Docker Compose, and Kubernetes?
● Docker Containers are a way to package software into a single unit
that can be run on any machine, regardless of its operating system. It
is used to create a Dockerfile, which is a text file that describes how to
build a Docker image. The image is then used to create a container,
which is a running instance of the image. This makes them ideal for
deploying applications on a variety of platforms, including cloud-based
services.
● Docker Compose is a tool provided by Docker that allows you to
define and manage multi-container Docker applications locally. It
enables you to use a YAML file to configure the services, networks,
and volumes needed for your application's setup. With Docker
Compose, you can describe the services your application requires,
their configurations, dependencies, and how they should interact with
each other, all in a single file. This makes it easier to orchestrate
complex applications locally composed of multiple interconnected
containers.
● Kubernetes is a container orchestration system that automates the
deployment, scaling, and management of containerized applications. It
allows you to run multiple containers on a single machine or across
multiple machines. It is an open source and can be deployed in your
data centre or the cloud.

16
10. What is the purpose of learning to develop APIs in a Generative AI
program?
APIs (Application Programming Interfaces) are used to connect different
software applications and services together. They are the building blocks of
the internet and are essential for the exchange of data between different
systems.

In the third quarter of the Certified Generative AI (GenEng) Developer and

Engineering Program, students will learn to develop APIs not just as a
backend for their front end but also as a product itself. In this model, the API
is at the core of the business's value.

● APIs are used to make it possible for different software applications to

communicate with each other.
● APIs are used to access data from a remote server.
● APIs are used to create new services or applications that are
integrated with existing systems.
● APIs are used to improve the security of applications by providing a
way to control access to data.
● By learning to develop APIs, students will gain the skills necessary to
create powerful and efficient software applications that can be used to
solve a variety of business problems.

11. What is the purpose of using Python-based FastAPI and related

technologies in Quarter 3?
In the third quarter of the Engineering Program, students will learn how to use
Python-based FastAPI as a core library for API development.

● FastAPI is a high-performance, lightweight, and easy-to-use framework

for building APIs.
● It is designed to be fast, scalable, and secure.
● FastAPI is compatible with a wide range of programming languages
and frameworks, making it a good choice for developers with different
skill sets.
● Students will also learn about the following related technologies:
● Pydantic: Pydantic is a Python library that helps to improve the quality
of your code by checking for errors and potential problems.
● SQLModel: SQLModel is a Python library that provides an object-
relational mapping (ORM) layer for working with databases.
● PostgreSQL: PostgreSQL is a free and open-source relational
database management system (RDBMS) that can be used for
development. Highly scalable database systems compatible with it
have also been deployed by all the major cloud platforms.

17
 By the end of the quarter, students will be able to use Python-based FastAPI
to develop APIs that are fast, scalable, and secure.

12. What does the API-as-a-Product model entail?

API-as-a-Product is a type of Software-as-a-Service that monetizes niche
functionality, typically served over HTTP. In this model, the API is at the core
of the business's value. The API-as-a-Product model is different from the
traditional API model, where APIs are used as a means to access data or
functionality from another application. In the API-as-a-Product model, the API
itself is the product that is being sold.

The benefits of the API-as-a-Product model include:

● Increased flexibility: APIs can be used to access data or functionality

from any application, regardless of the underlying platform or
technology. This gives businesses greater flexibility in how they
integrate APIs into their applications.
● Reduced development costs: APIs can be reused by multiple
applications, which can save businesses the time and expense of
developing their custom APIs.
● Improved scalability: APIs can be scaled up or down as needed,
which makes them well-suited for businesses with fluctuating or
unpredictable traffic demands.
● Enhanced security: APIs can be more secure than traditional
methods of data exchange, as they can be protected by a variety of
security measures, such as encryption and access control.

13. What are the benefits of using Docker Containers for development,
testing, and deployment?
Docker Containers are a fundamental building block for development, testing,
and deployment because they provide a consistent environment that can be
used across different systems. This eliminates the need to worry about
dependencies or compatibility issues, and it can help to improve the efficiency
of the development process. Additionally, Docker Containers can be used to
isolate applications, which can help to improve security and make it easier to
manage deployments.

14. What is the advantage of using open Docker, Kubernetes, and Terraform
technologies instead of using AWS, Azure, or Google Cloud
technologies?

18
Using open-source technologies like Docker, Kubernetes, and Terraform
offers several advantages over relying solely on proprietary cloud services
from AWS, Azure, or Google Cloud. Here’s a detailed comparison:

Advantages of Using Docker, Kubernetes, and Terraform (Open

Technologies)

1. Portability and Flexibility:

- Vendor Agnostic: These tools are cloud-agnostic, meaning you can run
your applications on any cloud provider or on-premises infrastructure without
being locked into a specific vendor.
- Ease of Migration: Applications packaged in Docker containers can easily
be moved across different environments, and Kubernetes provides a
consistent orchestration layer, ensuring seamless transitions.

2. Cost Efficiency:
- Avoid Vendor Lock-In: Being locked into a single cloud provider can lead
to higher costs over time. Using open technologies allows you to leverage
competitive pricing from multiple providers or even use on-premises
resources.
- Optimised Resource Utilisation: Kubernetes helps in efficiently managing
resources through automated scaling and load balancing, potentially reducing
costs.

3. Community and Ecosystem:

- Open Source: These tools are backed by strong, active open-source
communities that continuously improve the software, provide support, and
share best practices.
- Ecosystem: A rich ecosystem of tools and integrations is available,
providing flexibility to choose the best components that fit your specific needs.

4. Standardisation and Consistency:

- Unified Platform: Using Docker for containerization, Kubernetes for
orchestration, and Terraform for infrastructure as code (IaC) provides a
standardised way to deploy, manage, and scale applications across different
environments.
- Consistency Across Environments: These tools ensure that your
development, staging, and production environments are consistent, reducing
bugs and deployment issues.

5. Customization and Control:

- Full Control: Open-source tools give you complete control over your
infrastructure and deployment pipelines. You can customise and extend the
functionality to suit specific requirements.

19
 - Transparency: Access to the source code means you can audit and modify
the software to meet your security and compliance needs.

Advantages of Using AWS, Azure, or Google Cloud Technologies

1. Managed Services:
- **Ease of Use:** Cloud providers offer a wide range of managed services
that abstract away the complexity of setting up and managing infrastructure.
This can save time and reduce operational overhead.
- Integrated Solutions: These platforms provide integrated services and
tools, such as databases, machine learning, analytics, and monitoring, which
can be easily combined to build complex applications.

2. Scalability and Reliability:

- Global Infrastructure: Cloud providers have extensive global infrastructure,
ensuring high availability, redundancy, and low latency.
- Auto-Scaling: Advanced auto-scaling capabilities can dynamically adjust
resources to meet changing demands, ensuring optimal performance.

3. Security and Compliance:

- Built-In Security: Cloud providers offer robust security features, including
identity and access management, encryption, and compliance certifications,
helping to protect your data and meet regulatory requirements.
- Automatic Updates: Managed services often include automatic updates
and patches, reducing the risk of security vulnerabilities.

4. Innovation and Support:

- Cutting-Edge Technology: Major cloud providers continuously innovate
and introduce new services and features, allowing you to leverage the latest
technologies without significant investment.
- Support and SLA: Comprehensive support services and Service Level
Agreements (SLAs) ensure that you have access to expert help and
guaranteed uptime.

Conclusion

Choosing between open-source technologies like Docker, Kubernetes, and

Terraform versus proprietary cloud services from AWS, Azure, or Google
Cloud depends on your specific needs and priorities.

- Open Technologies: Offer portability, cost efficiency, customization, and

control, making them ideal for multi-cloud strategies, avoiding vendor lock-in,
and having more control over your infrastructure.

20
 - Cloud Providers: Provide ease of use, managed services, scalability,
security, and access to cutting-edge technology, which can be advantageous
for rapid development, scaling, and leveraging advanced services.

In many cases, a hybrid approach that combines the strengths of both open-
source tools and cloud provider services can provide the best of both worlds,
allowing you to optimise for cost, flexibility, and innovation.

15. Why in this program are we not learning to build LLMs ourselves? How
difficult is it to develop an LLM like ChatGPT 4 or Google’s Gemini?
Developing an LLM like ChatGPT 4 or Google Gemini is extremely difficult
and requires a complex combination of resources, expertise, and
infrastructure. Here's a breakdown of the key challenges:

Technical hurdles:

Massive data requirements: Training these models requires an immense

amount of high-quality data, often exceeding petabytes. Compiling, cleaning,
and structuring this data is a monumental task.
Computational power: Training LLMs demands incredible computational
resources, like high-performance GPUs and specialised AI hardware. Access
to these resources and the ability to optimise training processes are crucial.
Model architecture: Designing the LLM's architecture involves complex
decisions about parameters, layers, and attention mechanisms. Optimising
this architecture for performance and efficiency is critical.
Evaluation and bias: Evaluating the performance of LLMs involves diverse
benchmarks and careful monitoring for biases and harmful outputs. Mitigating
these biases is an ongoing research challenge.

Resource and expertise:

Team effort: Developing an LLM like ChatGPT 4 or Google Gemini requires a

large team of experts across various disciplines, including AI researchers,
machine learning engineers, data scientists, and software developers.
Financial investment: The financial resources needed are substantial,
covering costs for data acquisition, hardware, software, and talent. Access to
sustained funding is critical.

Additionally:

Ethical considerations: LLMs raise ethical concerns like potential misuse,

misinformation, and societal impacts. Responsible development and
deployment are crucial.

21
 Rapidly evolving field: The LLM landscape is constantly evolving, with new
research, models, and benchmarks emerging. Staying abreast of these
advancements is essential.

Therefore, while ChatGPT 4 and Google Gemini have made impressive

strides, developing similar LLMs remains a daunting task accessible only to a
handful of organisations with the necessary resources and expertise.

In simpler terms, it's like building a skyscraper of knowledge and intelligence.

You need the right materials (data), the right tools (hardware and software),
the right architects (experts), and a lot of hard work and attention to detail to
make it stand tall and function flawlessly.

Developing similar models would be a daunting task for individual developers

or smaller teams due to the enormous scale of resources and expertise
needed. However, as technology progresses and research findings become
more accessible, it might become incrementally more feasible for a broader
range of organisations or researchers to work on similar models, albeit at a
smaller scale or with fewer resources. At that time we might also start to focus
on developing LLMs ourselves.

To sum up, the focus of the program is not on LLM model development but on
applied Cloud GenAI Engineering (GenEng), application development, and
fine-tuning of foundational models. The program covers a wide range of topics
including Python, GenAI, Microserices, API, Database, Cloud Development,
and DevOps, which will give students a comprehensive understanding of
generative AI and prepare them for careers in this field.

16. Business wise does it make more sense to develop LLMs ourselves
from scratch or use LLMs developed by others and build applications
using these tools by using APIs and/or fine-tuning them?
Whether it makes more business sense to develop LLMs from scratch or
leverage existing ones through APIs and fine-tuning depends on several
factors specific to your situation. Here's a breakdown of the pros and cons to
help you decide:

Developing LLMs from scratch:

Pros:

Customization: You can tailor the LLM to your specific needs and data,
potentially achieving higher performance on relevant tasks.
Intellectual property: Owning the LLM allows you to claim intellectual property
rights and potentially monetize it through licensing or other means.

22
Control: You have full control over the training data, algorithms, and biases,
ensuring alignment with your ethical and business values.

Cons:

High cost: Building and training LLMs require significant technical expertise,
computational resources, and data, translating to high financial investment.
Time commitment: Developing an LLM is a time-consuming process,
potentially delaying your go-to-market with your application.
Technical expertise: You need a team of highly skilled AI specialists to
design, train, and maintain the LLM.

Using existing LLMs:

Pros:

Lower cost: Leveraging existing LLMs through APIs or fine-tuning is

significantly cheaper than building them from scratch.
Faster time to market: You can quickly integrate existing LLMs into your
applications, accelerating your launch timeline.
Reduced technical burden: You don't need a large team of AI specialists to
maintain the LLM itself

Cons:

Less customization: Existing LLMs are not specifically designed for your
needs, potentially leading to lower performance on some tasks.
Limited control: You rely on the data and biases of the existing LLM, which
might not align with your specific requirements.
Dependency on external parties: You are dependent on the availability and
maintenance of the LLM by its developers.

Here are some additional factors to consider:

The complexity of your application: Simpler applications might benefit

more from existing LLMs, while highly complex ones might require the
customization of a dedicated LLM.
Your available resources: If you have the financial and technical resources,
developing your own LLM might be feasible. Otherwise, existing options might
be more practical.
Your competitive landscape: If your competitors are using LLMs, you might
need to follow suit to remain competitive.

23
 Ultimately, the best decision depends on your specific needs, resources, and
business goals. Carefully evaluating the pros and cons of each approach will
help you choose the strategy that best aligns with your success.

17. What are Custom GPTs?

"Custom GPTs" refers to specialised versions of the Generative Pre-trained
Transformer (GPT) models that are tailored for specific tasks, industries, or
data types. These custom models are adapted from the base GPT
architecture, which is a type of language model developed by OpenAI.
Custom GPT models are trained or fine-tuned on specific datasets or for
particular applications, allowing them to perform better in those contexts
compared to the general-purpose models.

Here are some examples of what custom GPT models might be used for:

1. Industry-Specific Needs: A custom GPT for legal, medical, or financial

industries could be trained on domain-specific texts to understand and
generate industry-specific language more accurately.

2. Language and Localization: Models can be customised for different

languages or dialects that might not be well-represented in the training data of
the base model.

3. Company-Specific Applications: Organisations might develop a custom

GPT model trained on their own documents and communications to assist
with internal tasks like drafting emails, generating reports, or providing
customer support.

4. Educational Purposes: Educational institutions might develop custom

GPTs trained on educational material to assist in creating teaching materials
or providing tutoring in specific subjects.

5. Creative Writing and Entertainment: Custom models could be trained on

specific genres of literature or scripts to assist in creative writing or content
creation.

6. Technical and Scientific Research: A custom GPT model could be

trained on scientific literature to assist researchers in summarising papers,
generating hypotheses, or even drafting new research.

These custom models are created through a process of fine-tuning, where the
base GPT model is further trained (or 'fine-tuned') on a specific dataset. This
process allows the model to become more adept at understanding and
generating text that is relevant to the specific use case. Fine-tuning requires

24
 expertise in machine learning and natural language processing, as well as
access to relevant training data.

18. What are Actions in GPTs?

Actions are a way to connect custom GPTs to external APIs, allowing them to
access data or interact with the real-world. For example, you can use actions
to create a GPT that can book flights, send emails, or order pizza. Actions
are defined using the OpenAPI specification, which is a standard for
describing APIs. You can import an existing OpenAPI specification or create a
new one using the GPT editor.

19. What are AI Agents and how do they differ from Custom GPTs?
AI Agents and Custom GPTs are both tools that utilise artificial intelligence to
perform tasks, but they have distinct functionalities and use cases. Here’s a
breakdown of their differences:

AI Agents
AI Agents are autonomous programs that can perceive their environment,
make decisions, and act upon them to achieve specific goals. They often
interact with other systems or users, continuously learning and adapting
based on their experiences.

Key Characteristics:
1. Autonomy: AI Agents operate independently without continuous human
intervention.
2. Learning: They often employ machine learning algorithms to improve
performance over time.
3. Interactivity: AI Agents can interact with their environment, other systems,
and users.
4. Goal-Oriented: They are designed to achieve specific objectives and can
adapt their actions to optimise towards these goals.
5. Multi-Modal Capabilities: AI Agents can incorporate various forms of AI,
such as computer vision, natural language processing, and decision-making
algorithms.

Examples:
- Robotics: Autonomous robots that navigate and perform tasks.
- Virtual Assistants: Programs like Siri or Alexa that interact with users and
perform tasks based on voice commands.
- Game AI: Non-player characters (NPCs) that adapt and react to player
actions.

Custom GPTs

25
 Custom GPTs are tailored instances of OpenAI’s ChatGPT, launched in late
2022. They are designed for specific purposes and enhanced with context.
Each custom GPT can have a unique “personality,” including tone of voice,
language complexity, and responsiveness to specific topics. For example, a
financial institution’s custom GPT could be trained on financial reports and
industry-specific terminology, while a healthcare provider’s version might
focus on medical literature and health policy documents

Key Differences

1. Autonomy:
- AI Agents: Operate autonomously and continuously interact with their
environment.
- Custom GPTs: Typically respond to specific inputs and generate outputs
accordingly, but don’t operate autonomously beyond text generation tasks.

2. Learning and Adaptation:

- AI Agents: Often incorporate continuous learning and adaptation
mechanisms.
- Custom GPTs: Rely on pre-training and fine-tuning phases, with limited
continuous learning capabilities.

4. Interactivity:
- AI Agents: Can interact with both digital and physical environments.
- Custom GPTs: Primarily interact through text-based inputs and outputs.

In summary, while both AI Agents and Custom GPTs utilise AI, AI Agents are
designed for autonomous, goal-oriented actions in diverse environments, and
Custom GPTs are specialised in generating and understanding human-like
text for specific applications.

20. Do we need to use Design Thinking and BDD for designing custom
GPTs and AI Agents?
Design Thinking and Behavior-Driven Development (BDD) are methodologies
that can greatly enhance the process of designing custom GPTs and AI
Agents, though they are not strictly necessary. Here’s how each can be
beneficial:

Design Thinking
Design Thinking is a user-centred approach to innovation and problem-solving
that involves understanding the user, challenging assumptions, redefining
problems, and creating innovative solutions through iterative prototyping and
testing.

26
Benefits for Custom GPTs and AI Agents:
1. User-Centric Focus: Ensures that the AI solutions are tailored to the actual
needs and pain points of users.
2. Empathy: Helps in understanding the context and environment in which the
AI will be used, leading to more relevant and effective solutions.
3. Iterative Development: Encourages continuous testing and refinement of
ideas, leading to more robust and user-friendly AI models.
4. Collaboration: Promotes cross-disciplinary collaboration, which can bring
diverse perspectives and expertise to the design process.

Behaviour-Driven Development (BDD)

BDD is a software development methodology that encourages collaboration
between developers, QA, and non-technical stakeholders through the use of
natural language descriptions of the desired behaviour of the software.

Benefits for Custom GPTs and AI Agents:

1. Clear Requirements: Ensures that the requirements are clearly understood
and agreed upon by all stakeholders.
2. Testable Scenarios: Facilitates the creation of testable scenarios that can
validate the AI’s behaviour against the expected outcomes.
3. Documentation: Provides clear and comprehensive documentation of the
AI’s intended behaviour, which is useful for future maintenance and
enhancements.
4. Alignment: Ensures that the development stays aligned with business goals
and user expectations.

Application in Designing Custom GPTs and AI Agents

For Custom GPTs:

- Design Thinking:
- Understand the specific use cases and user interactions where the GPT
will be applied.
- Iterate on the model’s performance by gathering user feedback and refining
the fine-tuning process.
- Prototype different conversation flows and evaluate their effectiveness with
real users.

- BDD:
- Define the expected behaviours of the GPT in natural language scenarios.
- Create automated tests that validate the GPT’s responses against these
scenarios.

27
 - Ensure that the GPT’s behaviour aligns with user stories and business
requirements.

For AI Agents:
- Design Thinking:
- Map out the user journey and identify critical interaction points where the AI
Agent will provide value.
- Prototype and test the agent’s interactions in various environments to
ensure robustness and usability.
- Use empathy maps and personas to better understand and anticipate user
needs and behaviours.

- BDD:
- Write behaviour scenarios that describe how the AI Agent should react in
different situations.
- Develop tests that simulate these scenarios to verify the agent’s decision-
making and learning processes.
- Continuously refine the agent’s behaviour based on test results and user
feedback.

While not strictly necessary, Design Thinking and BDD can significantly
enhance the design and development process of custom GPTs and AI Agents
by ensuring a user-centred approach, clear requirements, and continuous
improvement through iterative testing and feedback. These methodologies
help in creating more effective, reliable, and user-friendly AI solutions.

21. When Fine-Tuning Open-Source Large Language Models, how is FastAI

and PyTorch important and what role do these libraries play?
Fine-tuning open-source Large Language Models (LLMs) like Meta LLaMA 3
involves adapting pre-trained models to specific tasks or domains by
continuing the training process on a smaller, task-specific dataset. FastAI and
PyTorch play crucial roles in this process due to their powerful features and
ease of use. Here's how they contribute:

PyTorch
PyTorch is an open-source deep learning framework developed by
Facebook’s AI Research lab. It provides a flexible and efficient platform for
building and training neural networks.

Importance in Fine-Tuning LLMs:

1. Dynamic Computation Graphs: PyTorch’s dynamic computation graph (also
known as define-by-run) allows for flexibility and ease of debugging. This is
particularly useful when experimenting with different model architectures and
training strategies.

28
2. Extensive Libraries and Tools: PyTorch has a wide range of libraries and
tools that facilitate various deep learning tasks, including natural language
processing (NLP). Libraries like Hugging Face’s Transformers are built on top
of PyTorch, providing pre-trained models and utilities for fine-tuning.
3. GPU Acceleration: PyTorch supports GPU acceleration, which is essential
for handling the large computational requirements of fine-tuning LLMs.
4. Community and Ecosystem: PyTorch has a strong community and
extensive documentation, making it easier to find resources, tutorials, and
support for fine-tuning tasks.

Role in Fine-Tuning:
- Model Customization: Allows users to modify the architecture of LLMs to
better fit specific tasks.
- Efficient Training: Provides efficient backpropagation and optimization
routines to train large models on specialised datasets.
- Experimentation: Facilitates rapid experimentation with different
hyperparameters and training setups due to its flexible framework.

FastAI
FastAI is a deep learning library built on top of PyTorch that aims to simplify
training neural networks by providing high-level abstractions and best
practices.

Importance in Fine-Tuning LLMs:

1. Ease of Use: FastAI provides high-level APIs that simplify many complex
tasks involved in training and fine-tuning models, making it accessible to both
beginners and experts.
2. State-of-the-Art Techniques: FastAI incorporates many state-of-the-art
techniques and best practices in deep learning, such as learning rate
schedules, data augmentation, and transfer learning, which can improve the
performance of fine-tuned models.
3. Rapid Prototyping: FastAI’s concise and readable code allows for rapid
prototyping and iteration, which is crucial for experimenting with different fine-
tuning approaches.
4. Integrated Data Handling: FastAI offers powerful data handling and
preprocessing utilities, which streamline the process of preparing datasets for
training.

Role in Fine-Tuning:
- High-Level API: Simplifies the process of loading pre-trained models,
defining custom datasets, and setting up training loops.

29
 - Training Utilities: Provides utilities for monitoring training progress, adjusting
learning rates, and saving model checkpoints.
- Best Practices: Encourages the use of best practices in model training, such
as using discriminative learning rates and employing effective data
augmentation techniques.

Combined Workflow for Fine-Tuning LLMs

1. Model Loading: Use PyTorch (and libraries like Hugging Face

Transformers) to load a pre-trained LLM.
2. Data Preparation: Leverage FastAI’s data handling capabilities to
preprocess and load the fine-tuning dataset.
3. Training Setup: Define the model architecture and training parameters
using PyTorch, and use FastAI’s high-level APIs to simplify this process.
4. Fine-Tuning: Utilise FastAI’s training utilities to fine-tune the model on the
specific dataset, adjusting hyperparameters as needed.
5. Evaluation and Iteration: Evaluate the fine-tuned model using PyTorch’s
evaluation utilities, and iterate on the fine-tuning process by leveraging
FastAI’s rapid prototyping capabilities.

Both PyTorch and FastAI play crucial roles in the fine-tuning of open-source
LLMs. PyTorch provides the foundational tools and flexibility needed for deep
learning, while FastAI builds on top of PyTorch to offer high-level abstractions
and best practices that streamline the fine-tuning process. Together, they
enable efficient, effective, and user-friendly fine-tuning of large language
models for specific tasks and domains.

22. In this course both PyTorch and FastAI play crucial roles in the fine-
tuning of open-source LLMs, why don't we use TensorFlow instead?
While TensorFlow is a powerful and widely-used deep learning framework,
PyTorch and FastAI offer several advantages that make them particularly
well-suited for fine-tuning open-source Large Language Models (LLMs).
Here’s a detailed comparison highlighting why PyTorch and FastAI might be
preferred over TensorFlow for this specific task:

PyTorch vs. TensorFlow

1. Dynamic Computation Graphs:

- PyTorch: Uses dynamic computation graphs (define-by-run), which allow
for greater flexibility and ease of debugging. This is especially useful when
experimenting with new models and training strategies.
- TensorFlow: Initially used static computation graphs (define-and-run).
Although TensorFlow 2.0 introduced eager execution to support dynamic

30
graphs, PyTorch's implementation is often considered more intuitive and
easier to work with for dynamic tasks.

2. Ease of Use:
- PyTorch: Known for its simplicity and clear, Pythonic code, which makes it
easier to learn and use, especially for research and prototyping.
- TensorFlow: While TensorFlow 2.0 improved usability, it is still considered
more complex compared to PyTorch, particularly for newcomers.

3. Community and Ecosystem:

- PyTorch: Has seen rapid adoption in the research community, leading to a
rich ecosystem of tools, libraries, and community support. Libraries like
Hugging Face’s Transformers are built primarily for PyTorch, offering
extensive support for LLMs.
- TensorFlow: Has a strong industrial presence and is widely used in
production environments. However, the research community has increasingly
favoured PyTorch.

4. Integration with Hugging Face:

- PyTorch: Hugging Face’s Transformers library, which is a go-to for working
with LLMs, is deeply integrated with PyTorch. This library provides pre-trained
models, tokenizers, and utilities that simplify the process of fine-tuning LLMs.
- TensorFlow: Although Hugging Face provides TensorFlow support, the
integration is not as seamless or feature-rich as it is with PyTorch.

FastAI vs. TensorFlow

1. High-Level Abstractions:
- FastAI: Provides high-level abstractions that simplify complex tasks in
model training and fine-tuning, making it accessible and efficient for users.
These abstractions are built on top of PyTorch, leveraging its flexibility.
- TensorFlow: While Keras (part of TensorFlow) offers high-level APIs,
FastAI’s APIs are often considered more intuitive and tailored towards rapid
experimentation and prototyping.

2. Best Practices:
- FastAI: Incorporates state-of-the-art techniques and best practices in deep
learning, such as learning rate schedules, transfer learning, and data
augmentation, making it easier to achieve high performance with minimal
effort.
- TensorFlow: Requires more manual effort to implement many of these
best practices, which can be a barrier for quick iteration and experimentation.

3. Community and Learning Resources:

31
 - FastAI: Has an active community and a wealth of educational resources,
including courses and documentation that focus on practical, hands-on
learning.
- TensorFlow: Also has extensive documentation and resources, but the
learning curve can be steeper, especially for those new to deep learning.

Summary of Why PyTorch and FastAI Might Be Preferred

- Flexibility and Debugging: PyTorch’s dynamic computation graph is more

flexible and easier to debug.
- User-Friendly: PyTorch and FastAI offer more intuitive and user-friendly
interfaces, making them suitable for rapid prototyping and experimentation.
- Research and Development: PyTorch is preferred in the research
community, leading to faster adoption of new techniques and tools like
Hugging Face’s Transformers.
- Seamless Integration: Hugging Face’s deep integration with PyTorch
provides a robust ecosystem for working with LLMs.
- Educational Resources: FastAI’s focus on best practices and practical
education makes it easier for users to get started and achieve good results
quickly.

While TensorFlow remains a powerful framework, particularly in production

environments, PyTorch and FastAI provide a combination of flexibility, ease of
use, and community support that make them particularly well-suited for the
fine-tuning of open-source LLMs.

23. What is Physical AI?

Physical AI refers to the integration of artificial intelligence with physical
entities, such as robots, that can operate and interact in the real world. This
concept involves AI systems that not only process data and make decisions
but also perform physical actions and understand the laws of physics.

Key Characteristics:

1. Real-World Interaction:
- Physical AI systems can perceive their environment through sensors,
process this information, and take appropriate actions using actuators.

2. Embodiment:
- Unlike purely digital AI, Physical AI involves AI embedded in physical
bodies, like humanoid robots, which can navigate and manipulate the physical
world.

3. Understanding Physics:

32
 - These AI systems are designed to comprehend and adhere to the physical
laws that govern real-world interactions, such as gravity, friction, and object
dynamics.

4. Human-like Functionality:
- Humanoid robots are a prime example of Physical AI, as they are built to
perform tasks in environments designed for humans, utilising a form factor
that mirrors human anatomy.

5. Data-Driven Training:
- Physical AI leverages vast amounts of real-world data to train AI models,
enabling robots to improve their performance through machine learning and
interaction experiences.

Applications:
- Healthcare:
- Assistive robots that help with patient care, rehabilitation, and surgery.
- Service Industry:
- Robots that perform tasks such as cleaning, delivery, and customer
service.
- Manufacturing:
- Industrial robots that assemble products, manage inventory, and ensure
quality control.
- Exploration:
- Robots designed for exploration in environments like space, underwater, or
disaster zones.

Physical AI represents a significant shift from traditional AI applications

confined to virtual environments. It aims to bridge the gap between digital
intelligence and physical capability, creating systems that can understand and
interact with the world in a human-like manner. This evolution has the
potential to revolutionise various industries by enhancing automation,
improving efficiency, and enabling new forms of human-machine
collaboration.

24. What are the different specialisations offered at the end of the program
and what are their benefits?
At the end of the GenEng certification program we offer six specialisations in
different fields:

Healthcare and Medical GenAI: This specialisation will teach students how
to use generative AI to improve healthcare and medical research. This is
relevant to fields such as drug discovery, personalised medicine, and surgery
planning.

33
Benefits:
● Learn how to use generative AI to identify diseases, develop new
drugs, and personalise treatment plans.
● Gain a deeper understanding of the ethical implications of using
generative AI in healthcare.
● Prepare for a career in a growing field with high demand for skilled
professionals.

Web3, Blockchain, and GenAI Integration: This specialisation will teach

students how to integrate generative AI with Web3 and blockchain
technologies. This is relevant to fields such as finance, healthcare, and supply
chain management.
Benefits:
● Learn how to create smart contracts and decentralised applications
(dApps).
● Gain a deeper understanding of the potential of blockchain technology
and how it can be used to improve business processes.
● Develop the skills necessary to work in a rapidly growing field with high
demand for skilled professionals.

Metaverse, 3D, and GenAI Integration: This specialisation will teach

students how to create and use 3D models and other immersive content
manually and with generative AI. This is relevant to fields such as gaming,
marketing, and architecture.
Benefits:
● Learn how to use generative AI to create realistic and immersive 3D
models.
● Develop the skills necessary to work in the growing field of virtual
reality (VR) and augmented reality (AR).
● Apply generative AI to solve real-world problems in areas such as
product design, marketing, and education.

GenAI for Accounting, Finance, and Banking: This specialisation will teach
students how to use generative AI to improve accounting, finance, and
banking processes. This is relevant to fields such as fraud detection, risk
management, and investment analysis.
Benefits:
● Learn how to use generative AI to automate tasks, identify patterns,
and make predictions.
● Gain a deeper understanding of the financial industry and how
generative AI can be used to improve its processes.

34
 ● Prepare for a career in a growing field with high demand for skilled
professionals.

GenAI for Engineers: This specialisation will teach students how to use
generative AI to improve engineering design and problem-solving. This is
relevant to fields such as manufacturing, construction, and product
development.
Benefits:
● Learn how to use generative AI to create simulations, optimize designs,
and predict failures.
● Gain a deeper understanding of the engineering design process and
how generative AI can be used to improve it.
● Prepare for a career in a growing field with high demand for skilled
professionals.

GenAI for Sales and Marketing: This specialisation will teach students how
to use generative AI to improve sales and marketing campaigns. This is
relevant to fields such as advertising, public relations, and customer service.
Benefits:
● Learn how to use generative AI to create personalised marketing
messages, generate leads, and track campaign performance.
● Gain a deeper understanding of the latest marketing trends and how
generative AI can be used to improve them.
● Prepare for a career in a growing field with high demand for skilled
professionals.

GenAI for Automation and Internet of Things (IoT):

● Provide Multi-Modal User Interface for the IoT systems: Multimodal
interaction exploits the synergic use of different modalities to optimise
the interactive tasks accomplished by the users. This allows a user to
use several input modes such as speech, touch, and visual to interact
with IoT systems.
● Improve efficiency and accuracy of industrial processes: By
implementing GenAI in automation and IoT systems, industries can
optimise their processes, reduce manual labour, and increase
productivity while ensuring higher accuracy and consistency.
● Enhance decision-making: GenAI can analyse vast amounts of data
collected by IoT sensors to derive valuable insights, enabling
businesses to make informed decisions regarding operations,
maintenance, and resource allocation.
● Personalise user experiences: GenAI can leverage IoT data to
understand user preferences and behaviours, enabling the creation of
personalised experiences across smart devices and IoT-enabled
systems.

35
GenAI for Cyber Security:
● Strengthen threat detection and response: GenAI can be used to
rapidly detect and respond to cyber threats by analysing large volumes
of security data in real time, identifying anomalies, and suggesting
appropriate countermeasures.
● Enhance security monitoring and analysis: GenAI can assist
security analysts in monitoring and analysing security logs, automating
threat detection, and providing insights into security risks and
vulnerabilities.
● Improve threat intelligence: GenAI can be used to gather and
analyse threat intelligence from various sources, enabling
organisations to stay informed about the latest threats and trends and
proactively strengthen their security posture.

36