At ShineTech, I architected a full-stack autonomous code generation platform that takes natural language requirements through to production-ready code. I built a LangGraph state-machine workflow for multi-file generation, using a file-tree architecture followed by iterative logic filling.
I also worked on an online education platform agent, optimizing latency with FastAPI asynchronous streaming. By applying summary memory to compress 32K history contexts, I reduced token costs by over 50%.
At Zhipu AI, I spearheaded backend architecture for the LLM Open Platform and built a high-performance MaaS gateway. I led the migration of business lines to Kubernetes and managed a backend team of 10+.
At Tencent, I architected the Microvision (Weishi) Feed recommendation engine and social interaction systems. I refactored legacy systems with Golang to support millions of DAU, and helped reduce latency to tens of milliseconds in the feed distribution engine.
