Industrial Internship Report

CloudThat Technologies Pvt. Ltd.

Submitted by

Maan Patel

(12102120601035)

In partial fulfillment for the award of the degree of

Bachelor of Technology
in
Artificial Intelligence and Data Science
( AI )

A.D. Patel Institute of Technology

Vallabh Vidyanagar

The Charutar Vidya Mandal (CVM) University,

Vallabh Vidyanagar - 388120

Jan-May 2025

q i
CVM University A. D. Patel Institute of Technology
 A.D. Patel Institute of Technology

BACHELOR OF TECHNOLOGY

(Artificial Intelligence and Data Science)

CERTIFICATE
This is to certify that Mr. Maan Patel (12102070601035) has submitted the Industrial
Internship report based on internship undergone at CloudThat Technologies Pvt. Ltd.
for a period of 24 weeks from 23/01/2025 to 23/07/2025 in partial fulfillment for the
degree of Bachelor of Technology in Artificial Intelligence and Data Science, A. D.
Patel Institute of Technology at The Charutar Vidya Mandal (CVM) University,
Vallabh Vidyanagar during the academic year 2024 – 25.

Prof. Mayur Ajmeri Dr. Dinesh J.

Prajapati
Assistant Professor, Professor and Head,
AI & DS Department, ADIT AI & DS Department,
ADIT

Date:14/05/2025

q ii
CVM University A. D. Patel Institute of Technology
Place: ADIT College

q iii
CVM University A. D. Patel Institute of Technology
 CAERTIFICATE

DECLARATION

I, Maan Patel (12102070601035), hereby declare that the Industrial Internship

report submitted in partial fulfillment for the degree of Bachelor of Engineering
in Food Processing Technology, A D Patel Institute of Technology, The
Charutar Vidya Mandal (CVM) University, Vallabh Vidyanagar, is a Bonafide
record of work carried out by me at CloudThat Technologies Pvt. Ltd. ,
under the supervision of Prof. Dr. D. J. Prajapati and that no part of this report
has been directly copied from any students’ reports or taken from any other
source, without providing due reference.

Name of Student Signature of Student

Maan Jayesh Patel _________________

q iv
CVM University A. D. Patel Institute of Technology
 Acknowledgement

The successful completion of any task is made possible through collaboration,

coordination, and the collective efforts of various knowledgeable individuals. I would like
to extend my heartfelt gratitude to Mr. Yerraballi Suresh Kumar Reddy (Lead), my
industry mentor, for his insightful guidance, continuous support, encouragement,
dedicated cooperation the course of our internship.

I am also sincerely thankful to Dr. Dinesh J. Prajapati, Head of the Department, and Dr.
Vishal N. Singh, Principal, for offering us this valuable internship opportunity through
the institute.

My special thanks to Prof. Mayur Ajmeri, my Internal Guide, for her constant support and
thoughtful advice during this journey. Lastly, I express my deep appreciation to all the
faculty members of the Department of Artificial Intelligence and Data Science, A. D.
Patel Institute of Technology, Vallabh Vidyanagar, for their guidance, encouragement,
and valuable suggestions.

Maan Jayesh Patel

1210212060103

q v
CVM University A. D. Patel Institute of Technology
 Abstract

This report provides a comprehensive overview of my internship experience,

centered on the design and development of two tools one is LLM (Large
Language Model) Cost Calculator platform. The project aimed to bridge user
interaction with real-time cost estimation for various LLM providers
available on amazon’s bedrock service through a structured, user interface.
Another is AI based AWS Architecture Diagram Generation, where user can
generate aws architectures by giving single prompt.

During the internship, I focused on designing the system architecture for a

scalable and modular solution, integrating technologies such as Streamlit for
the user interface and backend logic for cost computations. The application
allowed users to select or input model providers, regions, model names, and
token usage details, ultimately estimating costs using manually coded pricing
logic. The project also involved dynamic data scraping from the AWS
Bedrock pricing page and structured processing of that data. Also
implemented

This report discusses the goals, methodologies, tools, and outcomes of the
project while highlighting the technical challenges faced—such as ensuring
dynamic extraction of pricing data—and the strategies adopted to overcome
them.

q vi
CVM University A. D. Patel Institute of Technology
 List of Figures

Fig 5.1 Project Structure

Fig 5.2 Project flowchart
Fig 5.3 Home Page
Fig 5.4 LLM Calculator Interface
Fig 5.5 Result of LLM Calculator
Fig 5.6 Architecture Design Interface
Fig 5.7 Generated Diagram in editable canvas

q vii
CVM University A. D. Patel Institute of Technology
 List of Abbreviations

AI – Artificial Intelligence

LLM – Large Language Model

API – Application Programming Interface

IDE – Integrated Development Environment

UML- Unified Modelling Language

AWS – Amazon Web Service

UI – User interface

q viii
CVM University A. D. Patel Institute of Technology
 Table of Contents
Institute Certificate ……………………………………………………………...….. ii

Company Certificate …………………………………………………...…..……….. iii

Declaration …………………………………………………………………………...iv

Acknowledgement …………………………………………………...……….………v

Abstract ……………………………………………………………………….……...vi

List of Figures ……………………………………………………………….……....vii

List of Abbreviation…………………………………………………………………viii

Chapters

Chapter 1: Company Overview .................................................................................... 3

1.1 About CloudThat Technologies Pvt. Ltd.

1.2 Core Expertise
1.3 Technologies & Services

Chapter 2: Project Descriptions .................................................................................... 5

2.1 AWS Architecture Design Automation Tool

2.2 LLM Pricing Calculator for AWS Bedrock

Chapter 3: Technology Stack ........................................................................................6

3.1 Programming Languages

3.2 AI and Machine Learning Models
3.3 Libraries & Frameworks
3.4 Tools & Software
3.5 Cloud & Deployment Service

Chapter 4: System Analysis…………………………………………………………....8

4.1 Study of current system

4.2 Problem and Weakness of current system
4.3 Requirement of New System
4.4 System Visibility

1
CVM University
A D Patel Institute of Technology
 4.4.1 The System Contribution To Overall Objective
4.4.2 The System Implementation
4.5 Proposed System
4.6 Features of New System
4.7 Processes
4.8 Selection of Software, Methodology & Approach
4.8.1 List of Software Used to Develop the System
4.8.2 Agile Methodology Implementation

Chapter 5: System Design……….………………………...…………………………11

5.1 System Design & Methodology

5.1.1 Model, View & Controller
5.1.2 System Methodology
5.1.3 Agile Methodology
5.1.4 Agile Scrum
5.2 Structure Design
5.3 Project Architecture
5.3.1 Admin and User Flow Architecture
5.3.2 Sample of Forms, Reports & Interfaces

Chapter 6: Implementation………………………………..…………………………18

6.1 Module specification

6.2 Results

Chapter 7: Challenges & Solutions ..............................................................................20

4.1 Challenges in Architecture Design
4.2 Challenges in LLM Calculator

Chapter 8: Conclusion & Future Scope .......................................................................23

6.1 Conclusion
6.2 Future Scope & Enhancements

Chapter 9: References .................................................................................................25

9.1 AI Models & Libraries Documentation

9.2 Software Tools & APIs Used

2
CVM University
A D Patel Institute of Technology
 12102120601035
6

Chapter 1: Company Overview

1.1 About CloudThat Technologies Pvt. Ltd.:

CloudThat Technologies Pvt. Ltd. is a leading cloud training and consulting organization
committed to empowering enterprises and individuals through innovative cloud solutions
and upskilling programs. Founded in 2012, CloudThat has positioned itself as a pioneer in
the cloud enablement space, offering specialized services across platforms like Microsoft
Azure, Amazon Web Services (AWS), Google Cloud, and more.

The company offers a dual approach—providing enterprise-grade cloud consulting

services along with instructor-led training programs aimed at accelerating digital
transformation. With a strong focus on cloud architecture, DevOps, AI/ML solutions, and
data engineering, CloudThat serves clients across various domains, including IT, BFSI,
healthcare, manufacturing, and government sectors.

CloudThat has trained thousands of professionals globally and holds strategic

partnerships with major cloud providers, making it an authorized training partner for
Microsoft and AWS. The company’s consulting arm delivers end-to-end solutions,
including cloud migration, infrastructure setup, security, and managed services tailored to
client-specific needs.

Driven by a mission to "build a cloud-enabled world," CloudThat continues to foster

innovation and bridge the gap between academic knowledge and industry demands
through a blend of technical expertise and real-world application.

1.2 Core Expertise

 Cloud Consulting and Migration – Providing end-to-end cloud adoption strategies,
including cloud readiness assessment, infrastructure setup, and seamless migration
across AWS, Microsoft Azure, and Google Cloud platforms.

 Cloud Training and Certification – Delivering industry-recognized training

programs and certification courses for individuals and enterprises across cloud
technologies, DevOps, AI/ML, cybersecurity, and more.

 DevOps and Automation – Implementing CI/CD pipelines, Infrastructure as Code

(IaC), and automated cloud deployments to ensure rapid, reliable, and scalable
software delivery.

1
CVM University
A D Patel Institute of Technology
 12102120601035
6

 AI/ML and Data Engineering – Building intelligent data-driven solutions using

cloud-native AI/ML services, big data processing pipelines, and analytics
platforms.

 Managed Cloud Services – Offering continuous monitoring, optimization, and

support for cloud infrastructure to ensure high availability, performance, and cost-
efficiency.

 CloudThat Technologies continues to lead the digital transformation journey by

blending cloud expertise with cutting-edge innovation to empower businesses and
professionals worldwide.

2
CVM University
A D Patel Institute of Technology
 12102120601035
6

Chapter 2: Project Description

2.1 AWS Architecture Design Automation Tool

 Objective: Automate the generation of AWS architecture diagrams from user

inputs using serverless components.
 Implementation: Built a workflow using AWS Lambda, API Gateway, and
Amazon S3 to receive structured architectural input and return a dynamically
generated diagram via Mermaid.js syntax. The system accepts API requests,
processes architecture logic in Lambda, and returns visualized output through a
web interface.
 Challenges & Solutions: Addressed formatting inconsistencies by designing a
robust JSON schema for inputs and handled scalability by using a stateless
Lambda function to decouple logic from UI.

2.2 LLM Pricing Calculator for AWS Bedrock

 Objective: Build an interactive tool that estimates the cost of using LLMs (Large
Language Models) on AWS Bedrock, based on dynamic pricing and user-
specified parameters.
 Implementation: Developed a Streamlit-based conversational interface that
extracts provider, region, model, and token usage inputs from user queries. Used
Playwright for headless browser automation to scrape up-to-date pricing tables
from the AWS Bedrock console and structured them for efficient lookup.
 Challenges & Solutions: Managed dynamic tab-switching and dropdown parsing
on the AWS pricing page by programmatically triggering user actions in
Playwright and implemented error handling to manage incomplete or inconsistent
data.

3
CVM University
A D Patel Institute of Technology
 12102120601035
6

Chapter 3: Technology Stack

3.1 Programming Languages

• Python – Core language used for backend logic, web scraping automation, and
cost computation.
• Local java .jar server is used for converting Plant UML code into .png image
• Basic UI structuring in the Stream lit application.

3.2 LLM Models

• Claude sonnet 3 v1 from aws bedrock

3.3 Libraries & Frameworks

• Stream lit – Framework used to build the interactive user interface for the LLM
Cost Calculator.
• Playwright – Headless browser automation tool used to scrape live AWS Bedrock
pricing data.
• Plantuml – Used to dynamically render AWS architecture diagrams based on
puml inputs.
• Boto3 – Python SDK for interacting with AWS services like Lambda and
Bedrock.
• Fast API (Prototype) – Considered for lightweight REST API design in future
versions.

3.4 Tools & Software

• AWS Lambda – Serverless compute service for processing architectural data and
returning diagram outputs.
• Amazon API Gateway – Exposed RESTful endpoints to invoke Lambda functions
from client interfaces.
• VS Code – Primary IDE used for development and debugging.
• Git & GitHub – Version control and collaboration platform for project
management.

3.5 Cloud & Deployment Services

• Amazon Web Services (AWS) – Used for deploying architecture-related logic

through serverless components.
• Streamlit Sharing / Local Server – For deploying and testing the LLM Cost

4
CVM University
A D Patel Institute of Technology
 12102120601035
6

Calculator web application.

• Playwright CLI & Headless Chromium – For robust and scalable scraping across
dynamic AWS pricing interfaces.

Chapter 4: System Analysis

4.1. Study of Current System

Currently, AWS Bedrock users and cloud architects face two challenges:
1. LLM Pricing Complexity: AWS Bedrock supports multiple foundation models
(FMs) with varying token-based pricing structures. Users must manually navigate
pricing web page, dropdowns, and tables to estimate costs as per user’s
requirements, which is time-consuming and prone to human error.
2. Manual AWS Architecture Design: AWS solutions architects use third-party
diagramming tools or AWS-provided editors to design architecture. These tools
often require manual placement of components and lack intelligent, AI-driven
suggestions or auto-generation based on high-level intent.

4.2. Problem and Weakness of Current System

 Lack of automation in cost estimation for LLMs on AWS.

 Non-intuitive user experience for pricing exploration in Bedrock.
 Manual architecture diagramming is slow, error-prone, and disconnected from AI
workflows.
 No unified platform that provides interactive tools tailored for developers,
analysts, or business stakeholders looking to either plan costs or design AWS
solutions quickly.

4.3. Requirements of New System

The new system should:

 Automate pricing calculations for AWS Bedrock based on user inputs like token
and number of requests.
 Allow interactive, conversational or form-based estimation.
 Dynamically parse and use the latest AWS pricing data.
 Enable intelligent generation or editing of AWS architecture diagrams.
 Provide drag-and-drop UI components for architecture customization.
 Allow users to visually simulate cloud infrastructure based on intent or use-case.
 Be accessible via a simple web interface using Stream lit or a similar framework.

4.4. System Visibility

5
CVM University
A D Patel Institute of Technology
 12102120601035
6

4.4.1. The System’s Contribution to Overall Objective

These two tools contribute toward:
 Reducing decision-making time and improving cost transparency for ML
deployments on AWS.

 Simplifying architecture prototyping with minimal effort and technical know-how.

 Empowering users (students, professionals, startups) to work efficiently with
AWS Architecture Design.

4.4.2. System Implementation

 The LLM Calculator uses a backend scraper module to fetch and update pricing,
with a user-friendly frontend that supports both form-based input and dropdown
interactions.
 The Architecture Diagram Generator integrates an AI-driven suggestion engine
with AWS icons and drag-and-drop capabilities using Streamlit and PlantUML
library.

4.5. Proposed System

The proposed system combines two modules in one unified dashboard:
1. LLM Cost Estimator: Accepts structured or unstructured prompts and walks users
through model selection, region, and token count to calculate costs in real-time.
2. AI-Powered Diagram Generator: Generates AWS architecture diagrams either
from descriptions (e.g., “Build a scalable web app on AWS”) or manual edits,
with auto-alignment and AWS-compatible icons.

4.6. Features of New System

 Interactive UI with prompt/form-based inputs
 Live cost estimation based on token types and region
 Dynamic scraping of AWS Bedrock pricing data
 Add-on feature where user can calculate number of tokens by giving text or pdf.
 AI-assisted architecture generation with suggestions
 Drag-and-drop diagram editing
 Export functionality for diagrams and estimates
 Modular backend for extensibility

4.7. Processes
 Data scraping from AWS Bedrock pricing pages
 Conditional logic flow for estimating total cost
 Natural language processing to extract structured plant uml code
 Diagram rendering engine using AWS official icons
 Component placement algorithm for clean UI layout
 Real-time interaction through Stream lit app

6
CVM University
A D Patel Institute of Technology
 12102120601035
6

4.8. Methodology & Approach

Agile Methodology Implementation

The project followed an Agile development approach with the following iterations:
 Sprint 1: Requirement analysis and backend scraping setup.
 Sprint 2: UI development for calculator and diagram generator.
 Sprint 3: Integration and real-time interaction testing.
 Sprint 4: Final refinements, user feedback incorporation, and deployment.
Each sprint included sprint planning, development, testing, and review meetings to ensure
continuous improvement and faster delivery.

7
CVM University
A D Patel Institute of Technology
 12102120601035
6

Chapter 5: System Design

5.1. System Design & Methodology

The design of the system follows a modular and scalable approach, separating core
functionalities such as input processing, data retrieval, cost calculation, and visual
generation into manageable components. A Model-View-Controller (MVC) inspired
structure is adopted for clarity, maintainability, and separation of concerns.

5.1.1. Model, View & Controller

 Model:
Responsible for data calculation logic, connection between backend and LLM,
including:
o Scraping AWS Bedrock pricing data using playwright library.
o Structuring models, regions, and token costs.
o Parsing user input and calculating estimated costs.
o For architecture design, user’s prompt is sent to aws lambda through aws
api gateway
o Lambda function passes system prompt and user’s description using
bedrock service and in response it gets structured puml code
 View:
The frontend interface built with Stream lit:
o Interactive forms, dropdowns, and chatbot UI for LLM calculator.
o Drag-and-drop canvas and AI-based diagram generation for architecture
tool.
o Responsive design for real-time feedback and visualization.
 Controller:
Connects the Model and View:
o Manages user events, such as form submissions or button clicks.
o Triggers cost estimations or diagram generation based on inputs.
o Coordinates updates between frontend UI and backend logic.

5.1.2. System Methodology

The system design is data-driven and user-focused:

 Uses modular services to decouple functionality.
 Employs API-based architecture, where applicable, for reusability.
 Implements real-time processing for a smooth user experience.
 Uses streamlined UI/UX flow for guiding users from input to result.

5.1.3. Agile Methodology

Agile methodology was employed throughout the development to ensure:

8
CVM University
A D Patel Institute of Technology
 12102120601035
6

 Flexibility in accommodating user feedback.

 Continuous integration and testing.
 Incremental progress with measurable outcomes at each sprint.

Each week was treated as a sprint that involved planning, development, internal testing,
and retrospective evaluation. Tasks were tracked and reassigned based on complexity and
team input.
5.1.4. Agile Scrum

The project followed a Scrum-based agile model:

 Product Backlog: Documented features such as region selector, LLM cost logic,
drag-drop editor, etc.
 Sprint Planning: Held weekly to assign tasks for frontend, backend, or diagram
logic.
 Daily Standups: Informal progress reviews and blocker identification.
 Sprint Review: Team reviewed output and integrated refinements.
 Retrospective: Assessed what worked, what didn’t, and how to improve.
This approach enabled iterative development with high adaptability.

5.2. Structure Design

The system’s internal structure is modular and service-based:

 LLM Cost Calculator Module
o playwright_scraper.py – pulls and parses AWS Bedrock pricing.
o calculator.py – applies pricing logic based on input tokens.
o ui_.py – Gets input (token and requests) from user and provide Stream lit
frontend for calculator.
 AWS Architecture Generator Module
o lambda_function.py – Sends request to model with system prompt and
user prompt and in response receive structured puml code.
o ui.py – Receive puml code and using aws icons stored in local system
and .jar local server puml code is converted into png image.
o diagram_generator.py – To make diagram editable it is needed to convert
into embedded draw io compatible xml file.
o ui_diagram.py – drawio provides extension to create drag-and-drop canvas
in iframe for editing.
 Common Utilities
o Logging

9
CVM University
A D Patel Institute of Technology
 12102120601035
6

(Fig-5.1)

5.3. Project Architecture

The project is divided into two major user-facing modules, each catering to different
needs of developers, students, engineers, or cloud architects.

User (Developer / Student / Engineer):

 Interacts with a unified interface for both tools.
 Chooses between:
o Architecture Diagram Generator – to design or auto-generate cloud
architectures.
 Inputs required data through prompts or dropdown forms.
 Views, edits, and exports the results (cost or diagrams).

10
CVM University
A D Patel Institute of Technology
 12102120601035
6

5.3.1 Flow Diagram Sum (Fig – 5.2)

11
CVM University
A D Patel Institute of Technology
 12102120601035
6

5.3.2. Sample of Forms, Reports & Interfaces

LLM Calculator Interface:

 Model Provider Dropdown
 Region Dropdown
 Token Input Fields
 Result Display Table
 Prompt-based input option
Architecture Diagram Tool Interface:
 Prompt Textbox (e.g., “Deploy a secure 3-tier app”)
 Auto-generated Diagram Preview
 Editable canvas result into complete editable aws architecture diagram
 Component Toolbox (EC2, RDS, S3, VPC, etc.)
 Export option in multiple format (jpg/png/xml/json)
Screenshots:

Home Page selection between two tools

(Fig- 5.3)

12
CVM University
A D Patel Institute of Technology
 12102120601035
6

LLM Calculator Interface

(Fig- 5.4)

Screenshot of cost estimation output

(Fig – 5.5)

13
CVM University
A D Patel Institute of Technology
 12102120601035
6

AWS Architecture Design User Interface

(Fig – 5.6)

Screenshot of the generated AWS diagram

(Fig – 5.7)

CHAPTER 6: IMPLEMENTATION
14
CVM University
A D Patel Institute of Technology
 12102120601035
6

6.1. Module Specification

This chapter details the actual implementation of the core modules and user interfaces for
both tools developed as part of the project.

6.1.1. LLM Cost Estimator Module

 Objective: To calculate the estimated cost of using different LLM models on
AWS Bedrock based on user input such as model provider, region, input/output
token size, and number of requests.
 Core Features:
o Provider & Region selection
o Model-wise token pricing logic
o Interactive Stream lit UI with form-based input and dropdown menus
o Dynamic cost calculation and breakdown table
o Integration with web-scraped pricing data or predefined JSON

6.1.2. Diagram Generator Module

 Objective: To generate a cloud architecture diagram from a textual description

using AI.
 Core Features:
o Prompt input (e.g., “Build a secure 3-tier architecture with EC2, RDS, and
S3”)
o Sending this to LLM Claude Sonnet 3 v1 available on bedrock
o In response to system & user’s prompt LLM generates PUML code
o Mapping to AWS component icons from puml code using .jar local server
o Display png image to user on stream lit based front-end
o Provide editable option on canvas with drag-and-drop editing
o Canvas is extension of drawio presented in iframe on stream lit’s front end
o Multiple export options (png/jpg/xml)

6.1.3. Prompt Parser Component

 Functionality:
o Converts natural language prompts into structured fields (for both tools).
o Extracts relevant metadata: model type, number of tokens, service names,
etc.
o Enables automation of diagram creation and input form pre-fill.

6.1.4. Cost Calculation Engine

 Functionality:
o Retrieves token pricing based on provider, region, and model.
o Multiplies input/output tokens with request count and pricing factor.
o Supports tiered or fixed-rate models.

15
CVM University
A D Patel Institute of Technology
 12102120601035
6

o Returns total estimated cost with clear breakdown.

6.1.5. Diagram Rendering Engine

 Functionality:
o Maps identified AWS services to corresponding visual icons.
o Places them in a hierarchical layout representing architecture.
o Allows user modification through canvas.
o Option to save/export final design.

6.1.6. Stream lit UI Layer

 Functionality:
o Acts as the frontend interface for both tools.
o Allows switching between calculator and diagram generator.
o Contains:
 Dropdowns for model selection
 Input fields for tokens and region
 Prompt field for architecture input
 Export and Reset buttons
o Uses layout containers and session state for real-time interactivity.

6.1.7. Utility Functions & Data Handlers

 Data loading from scraped JSON

 Region mapping and default value handling
 Export of results (cost summary as text/CSV, diagrams as images)
 Logging user interactions (optional for debugging)

6.2 Results

The combined solution achieved the following key outcomes:

 A fully working LLM cost calculator that estimates cost of generative AI usage on
AWS Bedrock using minimal input and clear logic.
 A conversational UI that allows input through structured forms.
 An AI-based architecture diagram generator that transforms textual descriptions
into editable AWS diagrams using Large Language Model.
 A modern, responsive frontend using Stream lit, providing real-time user
interaction and results.
 The tools demonstrate potential use in cloud cost analysis, training, solution
design, and automated documentation.

16
CVM University
A D Patel Institute of Technology
 12102120601035
6

CHAPTER 7: CHALLENGES & SOLUTIONS

This chapter outlines the significant technical and design-related challenges encountered
during the development of both tools and how each was addressed through practical
solutions.

7.1 Challenges in Architecture Design

7.1.1. Limited to Specific AWS Services

Problem:
Initially, while using the Diagrams library, every AWS service component had to be
manually imported from the library. This restricted automation and made it difficult to
scale when new services were needed.

Solution:
To resolve this, we switched to the PlantUML diagramming approach. We downloaded
the complete AWS icon set locally and built a dynamic rendering logic. Now, the tool can
reference these icons directly and flexibly use them to create diagrams based on input
text, removing the need for hard-coded imports.

7.1.2. System Prompt Tuning for Output Clarity

Problem:
When generating architecture components from user prompts, the raw output from the
language model included unwanted text (like extra explanations or unrelated
suggestions), making it hard to parse cleanly for diagram generation.
Solution:
We designed and fine-tuned a system prompt that clearly instructed the model to return
structured output in a predefined JSON or YAML format. This reduced noise and made it
easier to map components accurately.

17
CVM University
A D Patel Institute of Technology
 12102120601035
6

7.1.3. Layout & Positioning of Diagram Components

Problem:
Random placement of components often led to unreadable or messy architecture
diagrams, especially when multiple services were involved.

Solution:
We introduced a basic grid-based lay outing logic to align and group services logically
(e.g., compute, storage, database layers). This significantly improved readability and
professional appearance of the generated diagrams.

7.2 Challenges in LLM Cost Estimator

7.2.1. Scraping and Maintaining Pricing Data
Problem:
The AWS Bedrock pricing page had dynamic tabs and dropdowns which made automated
scraping difficult and brittle when the structure changed.

Solution:
We implemented a dynamic scraping logic using Playwright, identifying tab structure and
dropdown dependencies in the HTML. The scraper now reliably extracts and maps
provider-region-model-token pricing from different sections.

7.2.2. Handling Missing or Partial Inputs

Problem:
Users often provide incomplete input (e.g., only model name without region or token
counts), causing failure in estimation or incorrect cost calculation.
Solution:
We implemented a multi-stage prompt parser and fallback mechanism. The tool now
either fills defaults or interactively asks follow-up questions (using Stream lit forms) to
ensure required data is gathered before estimation.

7.2.3. Region-Model-Provider Mapping Logic

18
CVM University
A D Patel Institute of Technology
 12102120601035
6

Problem:
Each model is not available in every region and pricing varies by both model and region,
making it hard to manage accurate calculations.
Solution:
We built a lookup dictionary that maps each model to its valid regions and pricing details.
If a user selects an unavailable combination, the system prompts them to select a valid
one or auto-corrects based on closest match.

7.2.4. Designing a Unified UI for Both Tools

Problem:
Combining both tools in a single Streamlit app while keeping the UI clean and user-
friendly was challenging, especially when both had different input types.
Solution:
We used Stream lit tabs and session states to modularize each tool's interface. This
allowed switching between the calculator and diagram generator smoothly while
maintaining a shared user experience design.

19
CVM University
A D Patel Institute of Technology
 12102120601035
6

CHAPTER 8: CONCLUSION & FUTURE SCOPE

8.1 Conclusion
The final semester project provided a valuable opportunity to design and develop two
impactful tools at the intersection of cloud computing, artificial intelligence, and user-
centric automation:
1. LLM Calculator for AWS Bedrock Services
2. AI-based AWS Architecture Diagram Generator

Both tools addressed real-world inefficiencies faced by developers, students, and

architects working with generative AI and AWS infrastructure. The LLM calculator
simplifies the cost estimation of running large language models by offering an intuitive
interface that parses user input, validates model-region combinations, and delivers
transparent pricing insights. Meanwhile, the AI architecture diagram generator enables
users to convert textual infrastructure descriptions into visually structured diagrams using
a combination of LLM, icon libraries, and layout logic.

Through iterative design, modular coding, and prompt engineering, the project achieved:
 Dynamic and interactive UI using Stream lit with stateful components.
 Accurate AWS pricing scraping and mapping across multiple providers, regions,
and models.
 Structured natural language parsing for cost estimation and diagram generation.
 Intelligent prompt-based layout rendering using Plant UML and AWS service
icons.
 Scalable backend logic for adding new services, regions, or diagram rules with
ease.

These achievements demonstrate the practical value of combining LLMs, backend logic,
and interface design to improve cloud-based planning and decision-making for a wide
range of users.

8.2 Future Scope & Enhancements

While the current version of both tools successfully delivers their core functionalities,

20
CVM University
A D Patel Institute of Technology
 12102120601035
6

several areas exist for future development and extension:

LLM Calculator Enhancements:
 Live integration with AWS APIs to ensure real-time accuracy of pricing and
model availability.
 Support for additional parameters, such as concurrency, input data size estimation,
or fine-tuning costs.
 Downloadable pricing reports for business proposals or internal budgeting.
 Chat-style assistant mode using a deployed LLM to guide users through cost
estimation interactively.

Architecture Diagram Generator Enhancements:

 Addition of other cloud providers (Azure, GCP) with icon sets and service
mappings.
 Drag-and-drop canvas editing for user customization after initial diagram
generation.
 Version control and saving of architecture diagrams linked to project names or
user sessions.
 AI validation of diagrams to recommend improvements or highlight missing
elements.
General Improvements:
 User authentication and history tracking to let users save, compare, and re-run tool
outputs.
 Integration into DevOps pipelines for auto-documentation of infrastructure or
model cost estimates.
 Deployment as SaaS with a dashboard-style interface, user onboarding, and admin
analytics (if expanded commercially).

21
CVM University
A D Patel Institute of Technology
 12102120601035
6

Chapter 9: References

8.1 Libraries Documentation

1. Playwright Docs - https://playwright.dev/python/docs/intro
2. Streamlit Docs - https://docs.streamlit.io/
3. Plantuml Docs - https://plantuml.com/guide
4. Diagrams Library - https://graphviz.org/docs/library/Graphviz2dra

8.2 Software Tools & APIs Used

1. Selenium WebDriver – https://www.selenium.dev/
2. Plantuml local web server - https://plantuml.com/download

22
CVM University
A D Patel Institute of Technology