Yu Hang L.

I build AI agents that reach into real systems, plus the MCP servers and integration work that make them useful past the demo. What I ship: MCP servers and agent tooling. Custom MCP servers that expose real capabilities to an LLM: code-graph queries over a repo, a pooled browser that many sessions share, project indexing, and multi-backend task dispatch behind one interface. Agent orchestration. Routing work across several model backends with retries, quota handling, and machine-checkable acceptance criteria, so a run either passes a real check or comes back for rework. No "the model said it worked." Integration and automation. REST API integration, Playwright against login-gated and JS-heavy sites, OCR pipelines, OpenXML document generation, scheduled jobs that stay up. Recent Upwork work, all rated 5.0: - AI agent development on Microsoft Foundry - OCR and computer-vision pipeline (Tesseract), $250 fixed - Coordinate-mapping and combination-filter engine in Python/NumPy, $350 fixed - OpenXML embedding for Word, $200 fixed - Browser automation reliability rescue on a scraper that kept dying My own stack, which is where most of this came from: a multi-model agent orchestrator with MCP integration, and a deterministic scoring engine written in Rust, exposed to Python through pyo3 and served behind an authenticated FastAPI layer. I care about correctness more than speed. Output that is wrong but looks fine is worse than an error, so I build in tests and reconciliation, and I will tell you upfront when an approach won't hold up rather than handing you something that breaks the first time it runs unattended. 100% Job Success, Top Rated, 17 jobs delivered. Fluent English, native Chinese.