AI / security engineer. Building the tooling layer for agentic development: observability, audit, safety, inference.
Currently shipping in the Claude Code ecosystem and the LLM security space. Most projects are framework-mapped (NIST CSF 2.0, ISO 27001:2022, OWASP ASVS 5.0, EU AI Act) and built defensive-first.
| Project | What it does |
|---|---|
| claude-logger | Tamper-proof audit logger for AI coding agents. SHA-256 hash chains, append-only JSONL, SIEM-ready. |
| claude-sonar | Voice-as-observability for Claude Code. Hear what every parallel session just did, without looking. |
- LeanLLM — layer-wise LLM inference reference implementation, runs Gemma 4 on 8 GB RAM.
- Witness — tamper-proof AI agent audit trail, ed25519-signed, EU AI Act / HIPAA / DPDP export packs.
- Tripwire — 63-probe arXiv-cited jailbreak detection harness with rule + LLM-judge composite scoring.
- Prism — OpenAI-compatible multi-LLM cost router with fallback and structured audit logging.
- Verdict — AI-native SOC platform: SIEM + SOAR replacement, attack-chain correlation, MITRE navigator.
claude-code-ecosystem · llm-safety · ai-observability · siem · soc-automation · compliance-ready-ai
PRs welcome on public repos. For research collabs or security disclosures, see individual repo SECURITY.md.