RAKSHAM - AI Cybersecurity Hackathon

RAKSHAM – AI Cybersecurity Hackathon IIT Delhi in collaboration with Amazon India now runs on UPI, digital banking, and wallets, and fraudsters are running on AI. Cloned voices of bank officials, deepfake videos of CEOs approving transfers, and AI-scripted scam workflows are hitting real people every day. RAKSHAM challenges student teams to design and build products that make India's digital financial ecosystem safer from AI-enabled threats. Organised by ARIES and DevClub, IIT Delhi , in collaboration with Amazon and supported by Learning Links Foundation . Prizes worth ₹4 Lakhs Problem Statements (choose one) 01.

Voice Cloning Fraud Detection: Design a safe, real-time way to identify voice-cloning fraud in banking or payment workflows, while protecting privacy and limiting disruption to legitimate users. 02.

AI-Driven Scam Pattern Recognition: Detect emerging AI-enabled scam workflows in UPI, banking, or wallet contexts and deliver timely, useful alerts without creating alert fatigue. 03. Deepfake Detection for

Financial Communications: Help people and institutions verify financial communications, detect manipulation, and create an actionable path for review, reporting, or response.

Rounds Round 1: Online Submission Submit a single

PDF containing: A pitch deck (max 10 slides, excluding an optional appendix) covering the problem, user, threat scenario, detection approach, validation plan, safeguards, deployment pathway, and 48-hour prototype plan An architectural overview (max 3 pages/slides) with an end-to-end system diagram, data flow, detection logic, human review and escalation path, and security/privacy mitigations Reviewers score only the submitted PDF, so it must stand on its own without a live pitch or external links.

Round 2: 48-Hour

Grand Finale: The top 30 teams build their prototype in person at IIT Delhi.

Evaluation (100 points): Problem fit & user understanding: 20 Technical architecture & detection logic: 20

Solution value & differentiation: 15 Feasibility & finale execution plan: 15 Safety, privacy & responsible AI: 15 Adoption, integration & scale: 10

Communication quality: 5 Substance matters more than slide polish. We reward products that are specific, buildable, and safe.

Eligibility: Open to students; teams of 2–4 members One submission per team

Rules: Use only synthetic, public, consented, or properly authorised data. No confidential bank, customer, or platform data. Disclose all AI-generated content, datasets, models, and third-party assets in an appendix. Incomplete submissions may be disqualified.

Important Dates

Registration deadline: 7th October 2026