About

Serre AI is an independent research lab focused on the formal foundations of AI reasoning. Named after Jean-Pierre Serre — Fields Medal, Abel Prize — whose work exemplifies the kind of rigorous, foundational thinking we aim to bring to AI.

Researcher

Oddur Sigurdsson

Based in Reykjavik. Background in software engineering and computational theory. Serre AI is a solo research operation — one person with AI agents as collaborators.

Methodology

Research is conducted by autonomous AI agents built on Claude Code. Agents read literature, design experiments, draft papers, and iterate on their own weaknesses — with strategic human oversight, not operational.

Every experiment is pre-registered. Claims are tracked in a knowledge graph with automated consistency checks. Multi-agent review catches errors before submission. The platform itself is a research artifact.

Research Areas

Computational Complexity of LLM Reasoning

Connecting empirical reasoning failures to formal complexity classes. When and why do models fail systematically?

Verification Theory

Characterizing when checking LLM outputs is computationally easier than generating them, and by how much.

Benchmark Design

Diagnostic tasks that isolate specific reasoning bottlenecks rather than measuring aggregate performance.