Haifa, Israel. I build open-source tooling that makes the work of AI agents verifiable: rules declared up front, and a replayable record of what each agent did.
| Project | Role | What it is |
|---|---|---|
| Bernstein | Maintainer | Open-source governance layer for AI agents: declarative rules, deterministic scheduling, a replayable run record. mcp.bernstein.run verifies run receipts. Apache-2.0 |
| TRACE | Contributor | Open attestation spec for agentic systems: reproducibility claim (§3.1.4), schema, conformance vectors. 10 merged pull requests, including a TypeScript verifier |
| CoSAI WS4 | Contributor, one open pull request | OASIS workstream on secure design patterns for agentic systems |
Most commits and pull requests in my activity are drafted by coding agents that I run. I answer for what is submitted under my name.
Smaller tools: segue handoff card for Claude Code compaction · brand-impersonation-detector brand impersonation detection with offline-verifiable evidence · tigra your LinkedIn export as a local network map · trilingual-retrieval-bench Russian, English and Hebrew retrieval corpus
Earlier projects (archived): voyant travel agent with fact checks · kotef coding agent · sentio RAG on LangGraph · meulex Slack RAG assistant · shafi legal question answering · RightLayout keyboard layout fixer for macOS