Registered user since Fri 6 May 2016
AI systems that can not only generate code but reason about it, prove it correct, and act on it autonomously. That’s what I’m building toward.
I’m Ferhat Erata, an Applied Scientist in the Automated Reasoning Group at Amazon Web Services (AWS). I work on neurosymbolic AI and multi-agentic autoformalization for code and reasoning, grounding large language models through reinforcement learning, theorem proving, and semantic uncertainty estimation. The goal: self-proving, agentic AI systems that can verify their own outputs.
I hold a PhD in Computer Science from Yale University, where I was advised by Ruzica Piskac and Jakub Szefer. My research explored the boundary between learning and logic, with a focus on security: investigating side-channel vulnerabilities in post-quantum cryptographic code, reverse-engineering quantum circuits from power traces, and building a probabilistic symbolic execution engine for secure computation.
Before Yale, I co-founded a software R&D startup applying formal methods in industry, contributing to European collaborative projects with Airbus, Daimler, and Ford. That experience grounded a conviction I still carry: formal reasoning belongs in real-world systems, not just papers.
Today, I’m passionate about agentic AI, neurosymbolic reasoning, and autoformalization. The common thread across everything I do is one question: how do we build AI that can reason over code, ground itself in logic, and operate safely, securely, and autonomously at scale
Contributions