Vasarlaavinash/blindspot-demo

🎯 Blindspot Demo

Live demo for the OpenEnv Hackathon "Blindspot" submission.

What this is: an interactive comparison of 3 policies (Random / Trending / SFT-trained Blindspot) on the unknown-unknowns discovery task.

Three modes:

• —🎓 Real user — pick one of 17 ML researchers; rewards are computed against held-out adoption.
• —🚀 Persona — 3 hand-crafted personas; the trained-policy responses are precomputed (cached) so you see the real SFT output instantly with zero GPU.
• —✍️ Your paragraph — paste a bio; the engine matches you to the closest user and runs the proxy policy.

Trained adapter: https://huggingface.co/Vasarlaavinash/blindspot-sft-1.5b Blog / writeup: Blog.md Code: https://github.com/vasarlalikhilavinash/blindspot-env