Bharat AI-SoC Challenge

Overview

The Ministry of Electronics & Information Technology (MeitY), through its Chips to Startup (C2S) Programme, in collaboration with Arm, is pleased to announce the launch of the second edition of the Bharat AI-SoC Challenge.

This nationwide challenge is led by Arm, a world-leading UK semiconductor design company with a strong presence in India, with IIT Delhi as the mentor institute. The initiative is supported by the UK Government through the UK-India Technology Security Initiative (TSI).

The inaugural Bharat AI-SoC Student Challenge, organised by Arm, MeitY C2S and IIT Delhi, attracted more than 3,000 students across 1,238 teams. Following three levels of evaluation and virtual mentorship, 150 shortlisted teams were narrowed to five winners.

The 2nd Edition expands eligibility nationwide to students and working professionals/developers. Participants compete on equal terms across five problem statements, progressing from Open Qualifying to Proof of Concept (PoC) Development.

What’s New in the 2nd Edition

Two Participant Categories

Students (UG/PG), Working Professionals & Developers compete on equal terms across five problem statements.

Two Challenge Phases

Open Qualifying submissions are due by 30 November 2026. Shortlisted teams then advance to PoC Development, with submissions due by 10 February 2027.

One Winner per Problem Statement

One winner for each problem statement is selected after the final live virtual demo and jury Q&A.

Technology Theme: Applied AI on SoC

Applied AI on SoC brings AI processing onto edge devices. Local processing can support real-time operation, privacy-aware use of data and offline capability, while teams consider performance, power and cost constraints.

The major shift in the 2nd Edition is from building AI applications to engineering AI-enabled products on SoCs.

Participants will explore how AI models integrate with processors, memory, accelerators and software stacks. The end-to-end engineering journey spans:

AI algorithm development
Optimisation
System integration
Edge deployment
Benchmarking
Validation
Product engineering

Challenge Format

Phase 1 — Open Qualifying

Submission deadline: 30 November 2026

Teams develop their initial solution and submit their report, source code and demo.

Phase 2 — PoC Development

Submission deadline: 10 February 2027

Shortlisted teams receive mentorship to develop and complete their proof of concept (PoC).

Project Problem Statements

All eligible participants may choose any one of the five problem statements, whether they are students or working professionals/developers.

PS1 — Offline crop-disease/pest detection and advisory for farmers Theme: Agriculture & Food Security

Objective

Develop an AI-powered assistant that helps farmers identify crop diseases, detect pests and receive actionable recommendations, running entirely on-device without internet connectivity.

Project Description

Teams will build an AI system capable of analysing crop images to:

Detect diseases and pests
Classify severity levels
Recommend treatment and intervention strategies
Provide region-specific advisory guidance
Support multiple local languages
Operate offline for rural communities with limited connectivity

Example Application Areas

Crop disease early warning
Pest identification and management
Fertiliser and irrigation advisory
Post-harvest loss reduction
Smallholder farmer decision support
Agri-extension worker tools

Hardware

Teams may use any Arm-based platform appropriate for their solution, including:

Arm-based Android smartphones
Raspberry Pi 5, CPU-only
Raspberry Pi 5 with AI HAT+ / AI HAT+ 2
Arm-based Linux devices

If a Hailo accelerator is used for vision workloads, vision models run through HailoRT on the Hailo NPU, while Llama runs through ExecuTorch on the Raspberry Pi Arm CPU. These are separate pipelines and should be clearly documented.

Required Technologies

Llama
ExecuTorch

Technical requirement: A model from the Llama family must be used for the advisory or reasoning component and deployed on-device using ExecuTorch as the reference runtime (for example, Llama 3.2 1B/3B-Instruct or a fine-tuned or distilled variant).

Optional Technologies

OpenCV
Hailo SDK
PlantVillage / Cassava / IP102 datasets
Speech processing frameworks for voice-based advisory
Multilingual NLP models

Deliverables

Source code
Application
Demonstration video
Technical report

PS2 — Smart recycling & waste-segregation assistant using multimodal interfaces Theme: Environment & Climate

Objective

Develop an AI-powered assistant that helps users identify, classify and properly dispose of waste materials using Llama and ExecuTorch running on Arm-based devices.

Project Description

Teams will develop an AI-powered assistant capable of:

Identifying waste categories
Providing disposal recommendations
Explaining recycling options
Promoting sustainable waste management practices
Educating users about environmental impact

Teams are free to design the most appropriate user experience for their solution.

Possible Interfaces

Android applications
Chat-based assistants
Voice-based assistants
Camera-enabled assistants
WhatsApp integrations
Multimodal AI experiences

Example Application Areas

Household recycling
E-waste management
Community recycling
Plastic waste reduction
Sustainability education

Hardware

Arm-based Android smartphones
Raspberry Pi 5, CPU-only
Raspberry Pi 5 with AI HAT+ / AI HAT+ 2
Arm-based Linux devices

If a Hailo accelerator is used for vision workloads, vision models run through HailoRT on the Hailo NPU, while Llama runs through ExecuTorch on the Raspberry Pi Arm CPU. These are separate pipelines and should be clearly documented.

Required Technologies

Llama
ExecuTorch

Technical requirement: A model from the Llama family must be used for the advisory or reasoning component and deployed on-device using ExecuTorch as the reference runtime.

Optional Technologies

OpenCV
Hailo SDK
Speech processing frameworks
Computer vision models

Functional Requirements

Classify waste materials
Provide disposal recommendations
Explain recycling options
Demonstrate use of Llama
Demonstrate use of ExecuTorch

Deliverables

Source code
Application or interface
Demonstration video
Technical report

PS3 — On-device threat, phishing & scam detection Theme: Cybersecurity & Digital Safety

Objective

Develop an on-device AI system that detects cybersecurity threats, including network intrusions, malware behaviour and phishing or scam communications, running entirely on Arm-based edge hardware without cloud dependency.

Project Description

Teams will build an AI-powered security assistant capable of:

Monitoring and classifying network traffic for anomalies and intrusions
Detecting phishing URLs
Detecting scam SMS messages or fraudulent communications
Explaining threats to non-expert users in plain language
Recommending remediation actions
Processing data locally on the device without relying on cloud services

Example Application Areas

Home and SME network intrusion detection
Scam call and SMS filtering
Phishing link detection
IoT device behaviour monitoring
Offline threat intelligence for air-gapped systems
Digital safety assistant for non-technical users

Hardware

Arm-based Android smartphones
Raspberry Pi 5, CPU-only
Raspberry Pi 5 with AI HAT+ / AI HAT+ 2
Arm-based Linux devices

If a Hailo accelerator is used for compatible workloads, it runs through HailoRT on the Hailo NPU, while Llama runs through ExecuTorch on the Raspberry Pi Arm CPU. These are separate pipelines and should be clearly documented.

Required Technologies

Llama
ExecuTorch

Technical requirement: A model from the Llama family must be used for the advisory or reasoning component and deployed on-device using ExecuTorch as the reference runtime.

Optional Technologies

CICIDS / UNSW-NB15 / NSL-KDD datasets
PhishTank / OpenPhish feeds
UCI SMS Spam Collection
Network packet capture libraries such as libpcap and Scapy
Hailo SDK

Deliverables

Source code
Application
Demonstration video
Technical report

PS4 — On-device personalised learning assistant Theme: Education

Objective

Develop an AI learning companion that creates personalised learning pathways based on learner goals, interests and current skill levels.

Project Description

Teams will create an AI assistant capable of:

Understanding learner goals
Identifying skill gaps
Recommending learning pathways
Generating study plans
Adapting recommendations over time

Example Application Areas

Career guidance
Skill development
Exam preparation
Professional certification pathways
Technical learning roadmaps
Lifelong learning plans

Required Technologies

Llama
ExecuTorch
Arm-based Android smartphone

Technical requirement: A model from the Llama family must be used for the advisory or reasoning component and deployed on-device using ExecuTorch as the reference runtime.

Deliverables

Source code
Application
Demonstration video
Technical report

PS5 — Real-time industrial defect detection & quality-control assistant Theme: Manufacturing & Industry 4.0

Objective

Develop an on-device AI system that performs real-time visual inspection of manufactured components, detects surface defects and anomalies, and assists quality control operators on the factory floor.

Project Description

Teams will build an AI-powered quality control assistant capable of:

Detecting surface defects, cracks and anomalies
Classifying defect type and severity
Alerting operators with actionable recommendations
Logging inspection results
Operating in real-time at the edge without cloud dependence

Example Application Areas

Printed circuit board inspection
Steel and metal surface defect detection
Textile and fabric quality control
Packaging integrity verification
Automotive component inspection
Pharmaceutical tablet and capsule inspection

Hardware

Raspberry Pi 5, CPU-only
Raspberry Pi 5 with AI HAT+ / AI HAT+ 2
Arm-based Linux devices
Arm-based Android devices with camera

If a Hailo accelerator is used for vision workloads, vision models run through HailoRT on the Hailo NPU, while Llama runs through ExecuTorch on the Raspberry Pi Arm CPU. These are separate pipelines and should be clearly documented.

Required Technologies

Llama
ExecuTorch

Technical requirement: A model from the Llama family must be used for the advisory or reasoning component and deployed on-device using ExecuTorch as the reference runtime.

Optional Technologies

OpenCV
Hailo SDK
MVTec AD / NEU Surface Defect / KolektorSDD datasets
Computer vision models such as YOLO and EfficientDet
Anomaly detection frameworks

Deliverables

Source code
Application
Demonstration video
Technical report

Eligibility Criteria

Participants must be Indian nationals.
Students must be nominated by their respective college.
Teams may have 1–3 members.
All members of a student team must be from the same college.
All members of a professional/developer team must be from the same organisation.
Every team must have a Team Leader.
Each student team must also have a Faculty Mentor.

Timeline

Milestone	Date	Activities
Launch	19 September 2026	The Bharat AI-SoC Challenge 2026–27 launches.
Registration closes	19 October 2026	Eligible participants must complete their registration by this date.
Phase 1 — Open Qualifying submission deadline	30 November 2026	Teams must submit their Phase 1 Open Qualifying project by this date.
Phase 1 shortlisted-team results	Mid-December 2026	Shortlisted teams will be announced and will advance to the PoC Development phase.
Phase 2 — PoC Development submission deadline	10 February 2027	Shortlisted teams must submit their proof of concept by this date.
Finals & Winners’ Announcement	Early March 2027	All finalists take part in the final evaluation, with one winner selected for each problem statement.

Challenge Guidelines & Terms

Only eligible students and working professionals/developers may participate.
Only the Team Leader should register the team.
Duplicate registrations may result in rejection or disqualification.
All registration details must be accurate.
Virtual mentoring will be provided by industry and academic experts.
Student teams may also receive local support from their college mentors.
Teams may contact mentors and experts by emailing support@armbharatchallenge.com.
The project must be hosted on GitHub and submitted using the challenge submission form.
Challenge-related notifications and updates will be posted on this page.
Winners must provide an appropriate college or organisation ID and supporting documents.
Winner selection is at the sole discretion of the challenge organisers.
Incorrect or duplicate entries may result in disqualification.

Support & Queries

For general support, email support@armbharatchallenge.com.

For technical questions relating to a problem statement, email tech-queries@armbharatchallenge.com and include the relevant problem statement number in the subject line.

Getting Started Resources are available to help teams begin their projects.

Challenge updates will be posted on this page.

From 1st Edition to the Launch of 2nd Edition of the Bharat AI-SoC Challenge

Rewards & Recognition

Recognition	Recipients	Reward
Winner	Each winning team	CASH AWARD FROM IIT DELHI
Certificates	All team members who submit completed projects	CERTIFICATE
Finalist Recognition	All finalists	BRANDED PROGRAMME MEMENTOES

Organising Partners

Category	Partner	Role
Government	MeitY & UK Government	MeitY is the Indian Government partner for this challenge. The initiative is supported by the UK Government through the UK-India Technology Security Initiative (TSI).
Academia	IIT Delhi	Lead academic mentor for the challenge
Industry	Arm (with support from Meta)	Industry mentorship, technical guidance and evaluation support

Registration

Students (UG/PG), Working Professionals & Developers are eligible to participate and compete on equal terms across all five problem statements.

Registration closes on 19 October 2026.

Original Work Requirements

Entries must:

Be the original work of the entrant(s)
Be solely owned by the entrant(s)
Not violate the intellectual property rights of any other person or entity

Complete the form below to register for the Bharat AI-SoC Challenge 2026–27.

Frequently Asked Questions

1. Who can participate?

Indian nationals who meet the requirements in the Eligibility Criteria section may participate.

2. How many people can be in a team?

Teams may have 1–3 members.

3. Can students from different institutions form a team?

No. All members of a student team must be from the same college.

4. Do student teams need a Faculty Mentor?

Yes. Each student team must have a Faculty Mentor.

5. Can professionals from different organisations form a team?

No. All members of a professional/developer team must be from the same organisation.

6. Can a team change its composition or challenge area?

One change of team composition or challenge area is permitted within 15 days of registration.

7. Can a team enter more than one problem statement?

No. A team may enter one of the five problem statements.

8. Who are the organisers of the Bharat AI-SoC Challenge?

The Bharat AI-SoC Challenge is led by Arm, with IIT Delhi as the mentor institute. MeitY is the Indian Government partner for this challenge, and the initiative is supported by the UK Government through the UK-India Technology Security Initiative (TSI). Arm participates with support from Meta.

9. Participant Materials and Intellectual Property

Participants are responsible for ensuring that they have all necessary rights, permissions and authorisations to use and submit any intellectual property, software, code, data, materials or other content used in connection with their participation in the Bharat AI-SoC Challenge ("Participant Materials").

Participants must not submit or disclose any confidential, proprietary or third-party information unless they are authorised to do so and such disclosure does not breach any confidentiality, employment, contractual or other obligation owed to any third party.

Participation in the Challenge does not require participants to disclose their own confidential information. Any Participant Materials voluntarily submitted or disclosed as part of the Challenge are provided at the participant's own discretion and responsibility.

Each participant remains responsible for ensuring that their participation and submission do not infringe, misappropriate or otherwise violate any third-party intellectual property or other rights.

Details of the 1st Edition of the Challenge: www.armbharatchallenge.com/edition1