Michele Joseph

I’m interested in AI reasoning stability and the systems that shape what humans and models trust. I’m building ideas around Compression-Aware Intelligence (CAI), Contradish, and the broader question of how we make intelligence more stable.

About

I work across AI and law-adjacent questions with a focus on reasoning stability and contradiction detection for modern models. I care about making invisible failures visible. I studied computer science and business at Cornell University and am an incoming J.D. candidate at Harvard Law School. Outside of work, I love to run, wander NYC, curate playlists, catch sunrise and sunsets, appreciate style/fashion, and have meaningful conversation.

Interests

AI reliability • model evaluation • reasoning stability • contradiction detection • law/regulation • philosophy of intelligence • running

Work

CAI Framework for understanding hallucination and reasoning instability through compression strain Contradish Testing layer for AI reasoning stability under equivalent inputs Contradiction Engineering Methods and tooling for surfacing contradictions in model reasoning GitHub Code and prototypes LinkedIn Background and experience Strava Where I log my runs