LLM Agent Developer · Shipping AI services to production · Contributor to DeerFlow, OpenAI Agents SDK & Google ADK
I build AI services end-to-end — from an empty repo to production.
- Zero → Production, repeatedly — Built, deployed, and operated AI services for real companies, from the first line of code through day-to-day operations.
- Beyond Python — Studied Spring to push past Python's backend limits, then migrated a FastAPI backend to Spring.
- Open Source — Shipping fixes to widely used AI agent frameworks.
| Project | Org | Merged | Focus |
|---|---|---|---|
DeerFlow  |
ByteDance |  |
Security hardening, distributed runtime, observability |
OpenAI Agents SDK  |
OpenAI |  |
MCP, run resumption, sandbox, tracing |
Google ADK  |
 |
Session state, memory | |
LangChain Azure  |
LangChain |  |
Azure AI tools |
LangChain Redis  |
LangChain |  |
Semantic cache, index migration |
RedisVL  |
Redis |  |
Hybrid search |
DeepTutor  |
HKUDS |  |
Runtime cancellation, LLM streaming |
Highlights
ByteDance · DeerFlow
- Closed two bypasses of the prompt-injection guardrail
- Made request trace IDs an end-to-end invariant
- Fixed multi-worker idempotent reuse blocking threads with 409s
- Stopped skill toggles from writing secrets to disk in plaintext
- Stopped checkpointing
view_imagebase64 payloads
OpenAI · openai-agents-python
- Kept MCP Streamable HTTP sessions alive through a transient 5xx
- Delivered tool-not-found output on server-managed resume
- Kept sandbox command paths POSIX on a Windows host
- Let tracing pick up an
OPENAI_API_KEYset after startup
Google · adk-python
- Stopped private model reasoning from reaching Memory Bank
- Stopped session rewind from wiping
create_sessionstate
LangChain · langchain-azure
LangChain · langchain-redis
Redis · redis-vl-python
HKUDS · DeepTutor
