ai research & development lab
AI systems that hold up in production.
Yoneda Labs designs agent infrastructure, data platforms, and research tooling for companies whose systems can't afford to guess — 20+ years in production, 122 published papers, 14 shipped products.
featured work
Systems in production
CMRPS
A structural classification of mathematical physics — 12 volumes, Lean-checked
The Classification of Mathematical Representations of Physical Systems: a twelve-volume series plus synthesis asking whether physics admits a structural classification analogous to the Classification of Finite Simple Groups. Kernel-checked in Lean 4 with 280 machine-checked claims, an executable Haskell taxonomy, and a Periodic Table of Mathematical Physics.
CatDB
A governed semantic data layer for AI agents
CatDB federates CRM, billing, and ERP systems behind a typed schema and refuses to answer questions it cannot prove — no hallucinated rows, no silent joins. Live demo federates a Postgres CRM and MySQL billing system behind one query interface.
ContextFS
Persistent memory for AI coding agents
A Model Context Protocol server and CLI giving Claude Code, Codex, and other agents cross-session recall. Hybrid keyword + semantic search over 2,300+ stored memories. Published on PyPI with a live marketing site.
YonedaRepo
Git-compatible source control for agents
A git-compatible repository layer purpose-built for AI agents: HTTP API + MCP server, server-side commits, no clone required. Full Git semantics implemented on Cloudflare Durable Objects with strong consistency.
AgentHero
Durable runtime + marketplace for multi-agent apps
An AI-agent marketplace platform built for a client: Next.js 16 + Supabase where Claude- and Codex-powered agents are deployed, supervised, and credit-metered. A companion Rust/Tokio runtime provides YAML-defined workflows, Git-worktree isolation, budgets, and crash recovery.
GrokRxiv
Agentic peer review for research papers
A multi-agent research pipeline that orchestrates parallel Claude research agents with Gemini peer review and Codex verification to autonomously draft, review, and publish formally-verified papers. 51 papers ingested; five fully-reviewed paper bundles published.
research library
122 papers, hosted here
Twenty-seven collections applying category theory to open problems in physics, mathematics, and AI. Every paper is a free PDF on this site.
Quantum Perspectivism & Foundational Problems
Physics & Quantum13 papersMathematical Representations of Quantum Gravity
Physics & Quantum12 papersTopological & Categorical Quantum Error Correction
Physics & Quantum11 papersDerived Functors & Teleological QEC
Physics & Quantum10 papersThe Joule Standard: Energy Economics of Intelligence
AI & Computation8 papersCondensed Representation Theory of Physics
Physics & Quantumclients & track record
Trusted in production
- Belden
- Zoro / Grainger
- TSI (Techno-Sciences / Orolia)
- Regis Corporation
- BetaCom
- Evident.io
- Pop Secret (ConAgra)
- Cogability
acquisition our engineering contributed to — Evident.io → Palo Alto Networks
uptime on a coast-guard-scale rescue coordination platform
package events processed daily with sub-second response
mission
Provable, or it doesn't ship.
An AI system that refuses to answer when it can't prove the answer is a feature, not a limitation. That standard runs through every product, paper, and engagement.