About
I'm Teng Li, a machine learning engineer at Zoom, based in the Seattle area.
I've been building software for about nine years, on both sides of the Pacific. I started at Alibaba.com in Hangzhou, working on systems that had to survive e-commerce scale. In 2020 I joined Zoom, first on middleware — the infrastructure layer other teams build on — then moved into ML: retrieval and RAG systems, and from there into agents. In March 2026 I moved from China to Seattle to work on what I think is the most interesting problem in the industry right now: AI agents that face real users.
Day to day that means designing the tools agents call, wiring up first- and third-party connectors and MCP servers, and building the eval harnesses that tell us whether any of it holds up outside a demo. The hard part is rarely the model — it's everything around it: the integrations, the failure modes, and the long tail of "spec-compliant but unusable."
What I write here
This site is where I think out loud. Three threads run through it:
- ML & Engineering — how production ML systems are actually built and kept alive.
- AI Agents & MCP — field notes on tool design, connectors, and evaluating agents that run against real APIs and real token bills.
- Career & Industry — the longer view on how an engineering career compounds across two countries, and what's shifting in a field that reinvents itself every few months.
I write to figure things out, and because the notes I most wanted to read — the ones from people running this stuff in production — were the ones nobody had written down.
Background
I studied at Northeastern University in China — a bachelor's in Biomedical Engineering, then a master's in Computer Application Technology. The pivot from biomedical signals to software stuck; so did the habit of treating systems like living things that fail in interesting ways.
Outside of work I'm usually reading, building side projects like mcpgrade, or spending time with my family.