Software that lets AItouch real systems safely.
Full-stack and AI systems engineer in Tel Aviv. Former CTO at Strongful, now building at Reptillix and on my own: MCP servers with real guardrails, agent platforms, and Hebrew-first products. TypeScript, Python, Go, Rust.
Available for new workTel Aviv, --:--
Selected work
More on GitHub
- secuservSafety self-test for your own MCP server
- llm-gatewayMulti-provider LLM gateway in Go with fallbacks and cost tracking
- rustcacheRedis-compatible cache in Rust, 50+ commands
- realtime-collabCollaborative editor on Yjs CRDTs with offline sync
- neural-searchHybrid dense and BM25 search with re-ranking, FastAPI
How I work
- Guardrails in the server, not the prompt
- Asking a model to behave is not a security control. Read-only connections, parsed queries, and redaction are enforced before anything reaches the database.
- Every claim has a test
- If a README says a thing is safe, there is a test file that proves it, and a map from each claim to that test.
- Hebrew first, not translated
- RTL is a layout decision, not a locale toggle. I build Hebrew products so the Hebrew version is the real one.
- Measure, then ship
- Cart abandonment down 15 percent, LCP down 40 percent, content time down 70 percent. Numbers, or it did not happen.
Now
- Building donation platforms and AI content pipelines for Reptillix clients.
- Pushing mcpolyglot out of alpha: write policies and more database drivers.
- Open to contract and full-time work.