Projects

Real-time AI fact-checker built in 24h for the Anthropic × IIT Madras CBC Hackathon. Streams live verdicts as users type via SSE: Claude Haiku for fast triage, Claude Sonnet for deep analysis. Won 1st Runner Up.

6-model ensemble (including an Audio Spectrogram Transformer) for multi-label music-genre classification. Hit 0.9614 Macro F1, reaching top 2.5% on the leaderboard. AST + SWA soups under tight compute constraints.

Memory-first AI mentor for career decisions, applications, and follow-through. Opens conversations with useful context from prior goals, patterns, commitments, documents, and past sessions.

A complete full-stack management app for tracking work, organizing operations, and moving day-to-day tasks through a clean production workflow. Built with a polished interface, responsive UX, and real shipped deployment.

RRDBNet built from scratch for the NPPE2 Plant-Leaves Super-Resolution challenge. 11-model ensemble with SWA model soups. Private leaderboard: score 16.569, rank 12 of 866 teams.