I'm Faaz Mohamed, founder of Adventura Labs, an independent AI research company I started in December 2025. I'm currently a sophomore at the University of Pittsburgh studying Computer Science and Data Science, and I'm one of five participants selected nationally for the NSF ExLAIM Research Internship.

My primary work is Project Coffeemaker, a multi-stage research program building a novel cognitive architecture called NOVA from scratch. Stage 0, a 307M parameter proof of mechanism, is complete. The core thesis is that epistemic honesty and capability are not in tension: safety properties like uncertainty awareness and architectural abstention should be baked into the forward pass, not bolted on afterward.

Before NOVA, I built LUMEN (a transformer with a mechanistic interpretability toolkit), VexDB (a C++20 vector database with SIMD/AVX2 acceleration), and VERITAS (a 9 microservice auditable AI evaluation system). Most of this was done during freshman year.

I care about interpretability, uncertainty quantification, and the question of whether safety can be structural rather than post-hoc. NOVA Stage 0 is my first attempt to build an empirical answer to that question.

Faaz Mohamed

Faaz Mohamed

NOVA Stage 0: Can Safety Be Structural? A Mechanism Proof at 307M Parameters