https://mihirm.com/ 2026-07-28 monthly 1.0 https://mihirm.com/og-image.jpg Mihir Mandavia — I build the thing first. Then I learn it inside out. https://mihirm.com/assets/portrait-clean.jpg Mihir Mandavia — portrait https://mihirm.com/assets/snitch-curves.png The Snitch — training reward, GRPO loss, policy KL, and eval reward curves from training an overseer model to catch reward-hacking AI agents https://mihirm.com/assets/snitch-team.jpg Team photo — Meta × Scaler OpenEnv Hackathon, where The Snitch placed in the Top 100 of 32,000+ entrants https://mihirm.com/assets/sih-team.jpg Team photo — Smart India Hackathon grand final, DualEDSR super-resolution project for ISRO thermal imagery