A LangGraph + Claude Agent SDK pipeline that reviews pull requests against security, licensing, and data-handling policy before merge — cost-gated, human-in-the-loop.
- active
- LangGraph
- Claude Agent SDK
- FastAPI
- Policy compliance
EE Heng Chen
Munich · CET
Built models that see. Now building agents that ship.
Computer-vision researcher (PhD & postdoc, 2018–2023) turned builder of LLM systems that turn regulation and review into structured, auditable output — deterministic decision layers, human-in-the-loop, evaluated.
Projects · 4
What I'm building
A pip-installable PDPA compliance layer for Malaysian SMEs using AI/LLM tooling with customer data. One shared codebase, two packages: presidio-audit and presidio-malaysia.
- active
- PDPA
- PII redaction
- Presidio
- Python
Multi-agent supervision of a live ML model: monitoring, investigation, remediation, post-mortem.
- active
- Multi-agent
- MLOps
- Monitoring
- Post-mortem
vcrpy for MCP. Records real agent–server sessions into diffable cassettes and replays them as deterministic mock servers.
- active
- MCP
- Testing
- PyPI
- Python
Recent writing · 2
Blog
Every fix to my incident loop after release was found and verified by reading code. When the demo finally ran end to end, it worked, and that changed what I think it's for.
The eval harness is the only signal for tuning the incident loop, which only works if ground truth stays hidden. A month after release I found postmortem memory handing it back.