PROBLEM STATEMENT 1:

AI/ML for Networking

Category: Network Security

Pre-requisites:

 Computer Systems Basics – CPU/Memory/Storage/NIC

 Good Hands-on Experience on Linux

 Programming Skills in Python and/or C

 Basics of AI/ML

Problem Statement

Description:

Modern networks face increasing challenges in monitoring and securing

traffic due to the exponential growth of data, encrypted communication,
and sophisticated cyber threats. Traditional rule-based security measures
and deep packet inspection (DPI) techniques are becoming less effective
in detecting and classifying threats, especially in encrypted traffic. Manual
intervention in network traffic classification is inefficient, leading to
delayed threat detection and security vulnerabilities. To address these
issues, AI-driven solutions can analyze traffic patterns, detect anomalies,
classify applications, and enhance security in real-time, ensuring adaptive
and intelligent network defense.

Expected Outcome:

 Automated Network Traffic Analysis using AI/ML models to

detect and classify traffic in real time.

 Improved Threat Detection & Security, identifying anomalies,

malware, and encrypted attacks with higher accuracy.

 Reduced False Positives & False Negatives, enhancing the

efficiency of network security operations.

 Scalability & Performance Optimization, ensuring AI models

can handle high-traffic environments with minimal latency.

 Privacy-Preserving Traffic Analysis, leveraging AI for encrypted

traffic analysis without decryption.
Deliverables:

1. AI-Powered Traffic Classification Model – A system that

categorizes network traffic (e.g., APP ID detection) based on
behavior and patterns.

2. Threat Detection & Anomaly Identification Framework – AI-

driven security mechanism to detect suspicious or malicious activity.
 PROBLEM STATEMENT 2:

Image Sharpening using knowledge distillation

Prerequisites:

Concepts in Machine Learning

Programming Skills (Python)

Deep Learning / CNN - Train/Validate/Test with Data

Objective:

Develop a model to enhance image sharpness during video conferencing,

addressing issues like reduced clarity due to low bandwidth or poor
internet connections.

Approach:

Utilize a Teacher-Student model technique for knowledge distillation:

Teacher Model: Select a high-performing pre-trained image sharpness

model.

Student Model: Design and develop an ultra-lightweight AI/ML model that

mimics the teacher model's performance.

Requirements:

The final model should operate at 30-60 frames per second (fps) or
higher, maintaining high accuracy.

During training, use high-resolution images cropped to lower resolutions

to reduce computational complexity. However, the model must be capable
of processing 1920x1080 resolution images at the target fps.

Training Dataset:

Downscale and upscale ground truth images using bicubic/bilinear

methods to simulate video conferencing conditions.

Performance Evaluation:

Use Structural Similarity Index (SSIM) as the performance metric.

Achieve an SSIM score above 90%.

Test the model on a benchmark dataset of over 100 images, including

diverse categories such as text, nature, people, animals, and games.
Conduct an extensive subjective study to obtain a Mean Opinion Score
(MOS) on the model's ability to produce sharpened images.

Expected Outcome:

A robust image sharpness model that enhances video conferencing

quality, providing clear and sharp visuals even under challenging network
conditions.

Report Submission Format :

Create a report citing data sources, detailed model(both teacher and

student) description, performance analysis process and result, and
working source code.
 PROBLEM STATEMENT 3:

Create pipeline (detect, decode and classification) using DL

Streamer, define system scalability for Intel HW
pre-requisites:
 Concepts in machine learning
 Programming skills (Python)
 OS (Linux)
Category: AI, machine learning, System scalability
Description:
As deployment of Edge and AI are growing, City and transportation are
adopting new use-cases where more and more visual cameras are
deploying across cites. It is very hard to manual scan all the cameras’
feeds. AI is helping to decode, detect and classify those cameras feeds
and providing analytics. Those cameras are used every where like
Mahakumbh or Cricket.
Example : https://www.livemint.com/news/mahakumbh-2025-how-ai-
powered-facial-recognition-is-assisting-law-enforcement-agencies-
surveillance-security-cctv-11737512022231.html
https://www.icc-cricket.com/media-releases/icc-tv-to-deliver-
comprehensive-coverage-with-ai-powered-innovations-and-star-studded-
commentary-team-for-icc-men-s-t20-world-cup-2024
Those AI cameras needs compute, storage and networking. Intel HWs are
the capable to handle those AI cameras.
DL streamer
(https://dlstreamer.github.io/dev_guide/dev_guide_index.html) optimized
to run at Intel HWs.
Create pipeline (decode, Detect and classify) on Intel HW (CPU and GPU)
run pipeline on CPU, GPU. Figure out how many cameras streams are
supported, what is optimum FPS which model best on Intel HW.
Outcomes:

 Maximum number of streams, FPS and Model on CPU

 Maximum number of streams, FPS and Model on GPU

 Bottleneck CPU, GPU, or IO

 Create a 3-page report on the chosen problem, Pipeline, and results

of each model
 PROBLEM STATEMENT 4:

AI-Powered Interactive Learning Assistant for Classrooms

Objective: Build a Multimodal AI assistant for classrooms to dynamically

answer queries using text, voice, and visuals while improving student
engagement with personalized responses.

Prerequisites:

 Familiarity with natural language processing (NLP) and multimodal

AI concepts.

 Knowledge of speech-to-text frameworks and computer vision

techniques.

 Programming skills in Python, with experience in libraries like

Hugging Face Transformers and OpenCV.

Problem Description:

Modern classrooms lack real-time, interactive tools to address diverse

student needs and keep them engaged. The objective is to create a
multimodal AI assistant that:

1. Accepts and processes text, voice, and visual queries from students
in real-time.

2. Provides contextual responses, including textual explanations,

charts, and visual aids.

3. Detects disengagement or confusion using facial expression analysis

and suggests interventions.

Expected Outcomes:

 A multimodal AI assistant capable of answering real-time queries

across various input formats.

 Integration of visual aids (e.g., diagrams, charts) for better

understanding.

 A feature to monitor student engagement and adapt teaching

methods dynamically.

Challenges Involved:

 Combining multimodal inputs (text, voice, visuals) for consistent,

context-aware responses.

 Ensuring low-latency processing to maintain real-time interactions.

  Handling diverse accents, noisy environments, and variations in
facial expressions.

Tools & Resources:

 Hardware: Intel AI PC with GPU and NPU for real-time processing /

any Intel Hardware.

 Software: Hugging Face Transformers (NLP), OpenCV (visual

analysis), PyTorch/TensorFlow.

 Datasets: Public multimodal datasets like AVA-Kinetics (for behavior

analysis) and LibriSpeech (for speech-to-text).
 PROBLEM STATEMENT 5:

Design and implement a smart, automated system for product

labeling and traceability, capable of verifying product quality
parameters and applying or validating labels that include:

 Device ID

 Batch ID

 Manufacturing Date

 RoHS Compliance

 Serial Number (QR or barcode)

 The system will leverage:

 Mechatronic systems for sensor-based inspection, actuation, and

real-time control.

 AI models for dynamic verification and validation of label data and

product compliance.

 Automated labeling or data printing mechanisms (simulated or real).

 Data logging for traceability and audit purposes.

Background:

 In modern manufacturing, especially in electronics, medical devices,

and automotive components, every product must be individually
labeled with traceable information to comply with regulatory,
logistical, and quality standards. This metadata allows tracking
through production, warehousing, and customer delivery.

 Currently, many factories use semi-automated labeling systems that

are error-prone or disconnected from quality checks. Your project
simulates the design of a smart labeling system that not only labels
the product but verifies critical parameters and ensures compliance
using sensor data and intelligent logic

Problem Description:

You are building a Smart Product Traceability Station in a manufacturing

line for small electronic devices.

System Requirements:

 Each unit arrives on a conveyor for inspection and labeling.

  The system must perform the following steps:

 Identify the product using a camera or sensor (or simulated input).

Verify compliance:

 Check RoHS status using part metadata or sensor flags.

 Confirm Batch ID from production input.

 Match Device ID and Manufacturing Date via database or simulation

logic.

 Apply or validate a label (with QR code, barcode, or digital record).

Use actuators to:

 Mark or print the label (real or simulated).

 Reject units with missing or mismatched data.

AI Integration:

 Use OCR (optical character recognition) or image recognition to

validate label print quality or content.

 Use machine learning to classify valid/invalid products or spot faulty

labels.

 Store all data into a simulated or real-time traceability log/database.

Deliverables:

 System Architecture Design:

 Block diagram, flowchart, and hardware/software mapping.

Label Inspection Simulation or Hardware:

 Simulated camera or sensor readings for product verification.

 Logic for generating and applying labels.

AI Module:

 OCR for label verification, or ML model for visual defect detection.

 Accuracy and performance report.

Control System Implementation:

 Embedded logic for sensor-actuator interaction and data validation.

 Label printing/rejecting mechanism.

 Traceability Database or File Logging:

  Store details like DeviceID, BatchID, timestamps, inspection result.

Final Demonstration and Report:

 Functional demo (video or live).

 Documentation covering design, implementation, dataset used, AI

training, and results.

Free Tools & Simulators for Students:

 Tool  Purpose  Link

 Simulate
 Tinkercad Circuits microcontroller and  Tinkercad
sensor logic

 Google Colab +
 Label inspection using
OpenCV +  Colab
AI/OCR in Python
EasyOCR

 LabelImg +  Train object detection

 YOLOv5
YOLOv5 in Colab model for label check

 Python +  Simulate traceability Python


SQLite/CSV database logging Docs

 Simulate control logic

 Proteus or Fritzing  Proteus
and label activation

 MATLAB +  Advanced modeling of MATLAB


Simulink label print systems Student

Evaluation Criteria:

 Scor
 Criteria  Description
e

 Application of sensors, actuators,

mechanical setup, and control
 1. Mechatronics logic for automated labeling and
 20
System Design inspection. System layout, design
clarity, and real-time execution
quality.

 2. Electronics &  Circuit design, microcontroller  15

Embedded programming, interfacing
Systems (sensors/actuators),
  Scor
 Criteria  Description
e

communication protocols
(I2C/SPI/UART), and power
management.

 Implementation and training of an

 3. AI and
OCR/ML model for label
Machine
verification or defect detection.  20
Learning
Evaluation metrics, model
Integration
integration, and performance.

 Accuracy and completeness of

 4. Data Logging logging product metadata (Device
& Traceability ID, Batch ID, etc.), use of  10
Logic databases or structured files, and
traceability relevance.

 Overall operation of the system,

 5. System including labeling accuracy, real-
Functionality & time rejection, product tracking,  10
Testing and system stability during
testing/demo.

 Creativity in design, optimization

techniques, edge computing, use
 6. Innovation &
of cloud or dashboard integration,  10
Problem-Solving
or other features extending the
baseline.

 7.  Quality of written report (clarity,

Documentation structure, diagrams), inclusion of
 5
& Technical results, challenges, and learning
Report outcomes.

 Clarity, professionalism,
 8. Presentation
teamwork, and effectiveness of
&  5
communication during the final
Demonstration
presentation/demo.

 Effective collaboration,
 9. Project
distribution of tasks, timeline
Management &  5
adherence, and group
Teamwork
coordination.

Dataset Recommendation for Defect Images

 For defect detection, especially in a real-world industrial scenario
like product labeling or quality control, using a relevant dataset will
help students train AI models effectively. Here are some datasets
that can be used for defect detection or quality inspection:

 The PCB Defect Detection Dataset on Kaggle:

 This dataset contains images of PCBs (Printed Circuit Boards) with

defects. It is perfect for students to apply computer vision
techniques for defect detection.

 Students can use OpenCV, TensorFlow, or PyTorch to implement

machine learning models for detecting defects in the products.

 Fruits and Vegetables Defect Dataset:

 This dataset includes images of fruits and vegetables with various

defects such as bruises, rot, etc. It’s a good fit for students to apply
defect detection models in food industry quality control systems.

 Steel Defect Dataset:

 This dataset contains images of steel sheets with different types of

surface defects (e.g., pits, cracks, etc.). It can be used for defect
detection in metal manufacturing processes.

 MVTec Anomaly Detection Dataset:

 MVTec provides a comprehensive dataset for industrial object

recognition and anomaly detection, with defects such as scratches,
dents, and missing parts on industrial objects. This dataset is
especially useful for students working with industrial automation in
visual inspection tasks.
 PROBLEM STATEMENT 6:

Product Category Creation for Healthcare Kiosks in India

Category: Artificial Intelligence, Machine Learning, Computer Vision,

Federated Learning, Healthcare, Electronic Health Records (EHRs),
Telemedicine,

Participants: Engineering Students or MBA Students with Engineering

background

Prerequisites:

 Understanding of Artificial Intelligence, Machine Learning, Computer

Vision, Federated Learning

 Understanding of Govt. of India policies and initiatives to transform

Indian healthcare industry, Digital India Mission, Ayushman Bharat
Digital Mission, National Health Database, Digital Public
Infrastructure, Digital Health Incentive Scheme (DHIS)

 Understanding of Product management to guide Intel on features

required through every step of a product’s lifecycle — from
development to positioning, competition mapping, market
requirements document and pricing — by focusing on the product
and its customers first and foremost.

Background – Healthcare Sector in India

The healthcare sector in India is one of the largest employers in the

country with 7.5 million people employed as of FY24. The Indian
healthcare market is currently valued at around US$ 638 billion.

The demand for Indian healthcare professionals is expected to double

nationally and globally by FY30 due to a shortage of healthcare workers in
India. India has 1.7 nurses per 1,000 people and a doctor-to-patient ratio
of 1:1500 nationwide which is worse than WHO standard recommendation
of doctor-to- patient ratio of 1:1000. India’s health expenditure increased
from 1.4% to 1.9% of the GDP from 2017-18 to 2023-24 which is still
below the target of 2.5% set by National Health Policy 2017. In actual
terms, the Indian government has allocated Rs. 99,858 crore (US$ 11.50
billion) to the healthcare sector in the Union Budget 2025-26 for the
development, maintenance, and enhancement of the country's healthcare
system.

India's healthcare landscape is undergoing a digital transformation, driven

by government initiatives, policy reforms, and technological
advancements. With a rapidly growing population and increasing demand
for quality healthcare, digital health solutions are playing a crucial role in
enhancing accessibility, affordability, and efficiency.

Digital healthcare infrastructure in India is evolving to bridge the gap

between urban and rural healthcare services, leveraging telemedicine,
electronic health records (EHRs), and artificial intelligence (AI)-driven
diagnostics. Govt. of India’s healthcare initiatives, such as the Ayushman
Bharat Digital Mission (ABDM) and the Digital Health Incentive Scheme
(DHIS), have the potential to set a global benchmark for digital healthcare
transformation. India is focusing on achieving the largest digitization drive
in the World by following basic principles like:

1. Interoperability and Standardization: Ensuring seamless data

exchange between key stakeholders

2. Data Management at Scale: Using the efforts put on creating the

India Stack which is a digital public infrastructure platform designed
to facilitate presence-less, paperless, and cashless services in India.

3. Public-Private Partnerships (PPP): Encouraging PPP model for quick

innovation, deployment and scale.

4. Focus on Affordability and Accessibility: Leveraging Govt. initiatives

like Digital India, Ayushman Bharat Digital Mission (ABDM), Digital
Health Incentive Scheme (DHIS), National Health Database (NHDB),
Ayushman Bharat Health Account (ABHA) and Pradhan Mantri
Ayushman Bharat Health Infrastructure Mission (PM-ABHIM) to
bridge the gap between Urban and Rural India in terms of
accessibility to quality health services while keeping affordability as
the focus.

5. Create differentiated delivery models: Indian Govt. is also creating a

comprehensive health service delivery mechanism spanning
preventive, promotive, curative, rehabilitative and palliative care.
There are two complementary components – Comprehensive
Primary Healthcare Units (Ayushman Arogya Mandirs) and
Comprehensive Health Insurance for poor populations (PM-JAY:
Pradhan Mantri Jan Arogya Yojana). Ayushman Arogya Mandirs are
envisaged to deliver expanded range services that go beyond
Maternal and child health care services to include care for non -
 communicable diseases, palliative and rehabilitative care, Oral, Eye
and ENT care, mental health and first level care for emergencies
and trauma, including free essential drugs and diagnostic services.
PM-JAY is the largest insurance scheme to cover more than 10 crore
Indian poor with a INR 5 lakh insurance cover to provide access to
secondary and tertiary care.

6. Vishwa Guru: India’s digital health models, with focus on leveraging

Artificial Intelligence, Machine Learning, Federated Learning and
Telemedicine, could serve as template for other developing nations
in the World to adopt and use for large scale Digital Public
Infrastructure in Healthcare.

Introduction – Healthcare Kiosk

A self-service kiosk has become a regular fixture in

many businesses irrespective of its size. Whether at
an airport or in a supermarket or in the hospitality,
banking, or healthcare sectors, this system makes
things easier for both customers and businesses.

The ability of healthcare kiosks to simplify processes

for patients and administrators offers great potential
to the healthcare industry. A healthcare kiosk is an
interactive device placed at a fixed spot in the hospital or
Ayushman Arogya Mandir, designed to accomplish multiple tasks that
reward patients and administrations alike.

A healthcare kiosk gives patients the control to find the information

themselves (wayfinding) which ensures that the patients are actively
engaged, as they can interact with the system in a meaningful way.
Current Features of the Healthcare Kiosk:

 Helps in booking walk in appointments.

 Expedite Patient Payments by integration of QR code-based

payments (UPI) or card payments

 Server as a ready reckoner for Indoor Patient Status

 Accurately measures BMI and patient vitals which are needed for
each patient before a doctor can access them during OPD visits

 Reduce Personnel and Material Costs

Technical Specification of the Healthcare Kiosk:

 Intel 12th Gen i3 / i5 / i7 based computing unit

 Touch Screen size 18.5”/22”/24”

 Ram 8gb/16gb

 Storage SSD 512gb

Problem Description

While a product on Intel Architecture exists, the off take is poor as it

solves certain market needs while ignoring many others.

Can the Healthcare Kiosk be improved to include certain workloads that

support the ABDM or DHIS or help by creating an innovative healthcare
delivery model or bring in AI, Machine learning and Federated learning to
bridge the urban-rural accessibility gap in India?

What features will unlock the potential to position the product as the right
product to solve the grassroot issues of India and create a product that
becomes synonymous with the principles of National Digital Health
Mission (NDHM) which aims to offer universal health coverage to all
Indians which unlocks a structured approach to interoperability,
cybersecurity, and data exchange mechanisms to ensure a robust digital
healthcare ecosystem creating a healthcare coverage that is efficient,
accessible, inclusive, affordable, timely and safe.

Can the Healthcare Kiosk be linked to the India Stack which is India’s
Digital Public Infrastructure platform comprises of various digital public
goods like UIDAI, eKYC, eSign, Digilocker, Unified Payments Interface, The
Account Aggregator framework, ONDC, Open Credit Enablement Network
(OCEN) and bring in Intel’s inherent secure assets like OneAPI, OpenVINO,
Open Edge Platform and Intel Product Security Assurance practices to
build a solution that consumers can inherently trust while sharing
confidential data under the NDHM?

Can technologies like face ID recognition be brought in with advanced

authorization from the patient, as seen in Digi Yatri Kiosks, to fasten the
registration and check-in process? If yes, what support is needed from the
ecosystem? Should we think of creating a foundation/body who will decide
on the features and its implications as well as implementation?

Can the Healthcare Kiosk serve as a federated learning platform across a

variety of machine learning model pipelines, a quality-preserving
database, an open benchmarking platform for comparing & validating AI
models to support diagnosis of non-communicable diseases, and a
consent management system for research of common cases under ABDM
thereby unlocking the immense potential of AI for improving health
outcomes?

Can the Healthcare Kiosk deployed at remote locations serve as a remote

consultations’ hub (tele-consultation), where patients can come in and
consult a doctor of choice located anywhere in the country thereby
reducing the burden on physical healthcare facilities? Can this be then
linked to the eSanjeevani (National Telemedicine Service) platform of
Govt. of India to serve as a system that immediately monitors patient's
vitals, sends a report along with medical history data with any relevant lab
or imaging results for the case being discussed by using AI and LLMs
(large language models) that too for Indians in various Indian Languages
and Dialects?

Can the Healthcare Kiosk self-propose a preliminary diagnosis and develop

an appropriate treatment plan which can then be taken to a doctor for a
second opinion before acting on the AI generated diagnosis and treatment
plan?

Can the Healthcare Kiosk be used to Connecting Health and Wellness

Centers (HWCs) or other hospitals with specialist doctors from anywhere
in the Country for tele-consultation for better healthcare accessibility in
remote areas or for sharing the tacit knowledge database? This Kiosk can
share the patient's vitals report along with medical history data with any
relevant lab or imaging results to the doctor located thousands of kms
away so that the doctor understands the history and recommends the
right treatment plan.
While all the above themes look interesting, we need to think about the
features that can be developed in todays’ World across the product
lifecycle, affordability in Indian context, funding sources to tap into, Go-to-
Market motions, the potential challenges in implementing a solution like
this in terms of regulations, product certifications, ground level
deployment and overall patient journey.

Expected Outcome:

 Mapping of current features and future features along with HW

requirements (current and future), SKUs, etc. based on the feature
introduction for the Healthcare Kiosk keeping in mind the 5 stages of
Product Management - (1) product development, (2) introduction,
(3) growth, (4) maturity, and (5) decline.

 Go-to-Market Blueprint for reaching target customers, building

demand, and gaining a competitive advantage keeping in mind the
various stakeholders like HW OEM, Independent Software Vendors
(ISVs) and System Integrators (SIs) as well as the customer support
mechanisms.

 Market requirements Document (MRD) to help define the market’s

requirements or demand for a specific product, typically containing
information on the product’s vision, the competitive landscape,
business analysis, and revenue opportunity, as well as a list of
features.