Senior software engineer, 11 years. Enterprise .NET systems, protocol reverse engineering, and AI tooling.
Designed approval-workflow and risk-scoring engines for financial institutions, and built the production systems behind a national F&B chain's store operations.
Open to senior platform / .NET roles, and to consulting on system rebuilds, protocol reverse engineering, and MCP agent infrastructure — tkflyc.com
Several AI CLIs run as background jobs behind one MCP interface. A registry architecture keeps each backend to a single file, and a circuit breaker trips a failing backend out instead of letting it block the others.
→ ai-cli-mcp-source · TypeScript · MCP SDK · node-pty
A Clean/Hexagonal .NET 10 server reconstructed from wire captures. Each protocol handler is labeled source-inferred, capture-verified, or live-verified, so inference and proof stay distinguishable in the docs instead of blurring together. Roslyn analyzers stop the core from referencing version-specific assemblies; 750+ tests; golden-fixture packet decoding against real captures.
→ MapleForge · C# .NET 10 · DDD · xUnit
A self-hosted knowledge server does hybrid search — FTS5 for lexical, BGE-M3 for vectors, RRF to fuse the two rankings. The public repo here is the client for it: it keeps no knowledge on the local machine, signs each request, and leaves the single copy on the server.
→ knowledge-mcp-client · Python · signed HTTP · MCP
-
Modular monolith, .NET 10 + Next.js 15 (private) — 14 module contracts checked by architecture tests, so a boundary violation fails the build instead of waiting for a reviewer to notice.
-
Knowledge server (private; the client above is open) — I run it as my own daily working memory, which is also how its retrieval quality gets tested.
-
Production monitoring (private) — converging alert thresholds. Working rule: when the monitoring itself breaks, it should be as loud as the service breaking.
Primary: C# / .NET 8–10 · TypeScript · Python
Reverse engineering: .NET decompilation · network protocol analysis · memory instrumentation
AI infrastructure: MCP SDK · hybrid retrieval (FTS5 + BGE-M3 + RRF) · ONNX