Final-year B.Tech CSE (AI & ML) at SRM, Chennai. Graduating 2027.
I work on making LLM systems less wrong, and on the tests that show whether they are. My usual approach is to put the logic that matters in plain code and use the model only for the messy parts.
Portfolio · Résumé · LinkedIn · airasalish@gmail.com
Warrant: an agent that reviews chargeback disputes and cites the evidence behind each decision. On a held-out set it made no wrong committed calls, and it held up against all 34 prompt-injection test cases. The README lists where it still falls short. Razorpay AI Buildathon 2026.
Receipts: finds stock tips in Indian finance YouTube videos, pins each to the second it was said, and scores how the stock did over the next 30 days against the Nifty 50. SerpApi India Hackathon 2026. Live
Drift: a watchlist that shows what changed since you last looked, using five rules that each show their evidence. 121 tests. Code, by Groww 2026. Live
Traverse: a FastAPI service that measures area and length from Shapefile and KML files, accurate to about one part in a billion against geodesic ground truth. Live
AIRA: turns a plain-English goal into a marketing campaign: picks the audience, drafts messages with Llama 3.3, and tracks delivery live. Live
HackerRank Orchestrate (24-hour AI agent hackathon): damage-claim verifier (June, #206), WhatsApp notification router (August, #27 of 1,983), Buy or Wait? (September, #84).
- Generative AI Researcher, BeSuperhuman.ai (Dec 2025 – Feb 2026). Built the release tests; wrong answers dropped 35%.
- Prompt Engineering & Automation Intern, BabyBillion (May – Sep 2025). AI video pipelines and Python automation.
- Technical Lead, AI/ML, ACM-W SRMIST (2023 – present).
AWS Certified AI Practitioner · SAP Certified Back-End Developer (ABAP Cloud) · Red Hat RH294 · Citi Bridge 2026 (top 5%) · GirlScript Summer of Code contributor (2025, 2026)
Mostly Python, TypeScript, FastAPI, Next.js, PostgreSQL and AWS.