a letter to Jane — simulation copy
I build the loop,
not just the features in it.
The panel beside this letter is my own agent, answering from my corpus under the scope this link opens. Ask it anything the letter leaves out — my background, my other companies, my work eligibility — while you read. On a phone it is the button at the bottom of the screen.
Dear Jane hiring team,
I am writing for the Senior Developer — Developer Acceleration seat. For the past couple of years I have been building StandMeet, a self-hosted platform where an AI agent works with a personal knowledge corpus. The parts of that product I spent myself on are the parts this role names: the agentic context and harness, the skills system other people build on, and the evals that decide whether an agent's work is actually good.
I started in security at a large telecom company, building network security components used across several product lines. Since then I have co-founded two small operating companies — FlexMesh in Canada and YouTeacher, registered in Wyoming — where I hold shares alongside my co-founders, and I build a reading app, Lucerna, with a partner. Most of the day-to-day operation of those projects is automated now; I step in once or twice a day to give direction. A full-time role would come first, and I am happy to disclose all of it.
What I am looking for is the part I cannot give myself: colleagues who are proactive — who bring their own ideas and their own questions. Given a good question, an agent's answer can be better than a human's; having the good question is the hard part, and it is the part I have been carrying alone.
Your posting, against my record
Agentic context and harness for ~300 developers
StandMeet is an agent platform I designed and built alone: the harness, the context system, the skills layer and the eval loop are the product. I have been running that harness for an audience of one — me. This role is the same job at the scale of a whole engineering org.
Skills and plugins, distributed safely
Skills are first-class in StandMeet: an owner MCP surface of 250+ tools, per-code corpus scope, sandboxed plugins (bubblewrap, no network, one unix socket back), and gates that reject changes an agent should not have made.
Evals and measurement, not vibes
My eval harness imports only the production agentcore facade and runs the byte-for-byte production loop against fixtures injected as legitimate data sources — grounding, privacy canaries, tool use, compaction past 32K, cross-conversation coherence. Quality is established by repeated empirical runs, not by reading prompts.
TypeScript across the stack
TypeScript services in YouTeacher and Lucerna, a Next.js/React front end on StandMeet, Go where the product needs it. I pick the language per surface and ship in all of them.
Platform scale and reliability
My own estate is single-host by design (Coolify + Traefik, one Postgres primary per product), with an incident record I keep honestly: upgrade races fixed with a version handshake and SKIP LOCKED leases, every fix proven by a test that failed on the old code first.
Honest gap: No multi-node HA in production, and a 300-developer org is new territory for me. I reason about both plainly rather than pretend otherwise.
Sijie Wang · Markham, Ontario · authorized to work in Canada. This page is a simulation built for rehearsal; the agent beside it is live.