Written-up delivery work, with the constraints and the numbers. Client names, contracts, customer data and internal endpoints are intentionally omitted; everything here is either my own architecture description or a sanitised metric.
Contact · cieve94107@gmail.com · LinkedIn · GitHub
| # | Case study | What it shows |
|---|---|---|
| 1 | Video surveillance & vehicle-device platform | GB28181/GA-T 1400 integration, live + playback streaming, multi-tenant data scope, production delivery discipline (tests, rollback, handover) |
| 2 | WeChat marketing & payments platform | Full WeChat Pay V3 loop, hot-configurable campaign engine, AI-assisted delivery (~80% of code), read-only MCP tooling for operations |
| 3 | mysql-ops-mcp (live repo) | Read-only-first MCP server: SSH tunnel management, SQL whitelist guard, mutating tools off by default, 24 unit tests |
- Deliver end to end, not a slice: requirements → architecture → implementation → deployment → documentation → source handover.
- Measure before claiming: test counts, stream counts, rollback drills. If something could not be verified on real hardware, it is written down as unverified instead of quietly assumed.
- AI in the loop, evidence in the repo: Claude Code / MCP servers / agent skills for speed, with the reasoning, decisions and verification steps kept as files so the next person (or agent) can pick it up.