Hi, I'm Deven. I work on AI safety and alignment, especially mechanistic interpretability.
Work
- Loyal Lies — 5th place / 179, Apart Research Secret Loyalties Hackathon 2026. Find out here · GitHub
- Crosswise — checks natural-language activation explanations against independent SAE evidence to catch fabricated claims. GitHub · Blog
- AuditAI — caste-bias audit of Gemma-2-2B using byte-identical counterfactual surname pairs. GitHub
- When Sentiment Circuits Don't Cross Borders — where sentiment computation lives in mBERT and why it doesn't transfer.
Experience
- Software Engineering Intern, Cryptographic Provenance Infrastructure — Authentic Memory (GAMI), Munich · Aug–Dec 2026
devensodyssey@gmail.com
Last Updated : 16 Aug'26