Juan L.

I take on engineering that the obvious options failed at. Reverse engineering undocumented APIs, running private LLM infrastructure on your own hardware, rewriting legacy backends, and rebuilding Codex, Claude Code, Cursor, Manus, Lovable, Bolt, or v0 prototypes as production-grade systems. Reverse engineering and undocumented API work: I turn black-box systems into structured, documented APIs. Recent work includes reverse-engineering a Flutter mobile app's auth flow and endpoint mapping for a TechCrunch-featured platform with 2.5M+ users, HAR-driven recon of multi-stage e-commerce order-capture flows, and building my own queryable database toolchain for HTTP traffic analysis (HAR/SAZ indexing, token tracing across captures, AI-assisted endpoint discovery). LLM and Agent AI infrastructure: I run and deploy production AI agents on local hardware, VPS, and self-hosted multi-GPU inference. Stack: llama-swap per-node manager, cluster router with FIFO queue, WhisperX transcription with diarization and voice-fingerprint enrollment, OpenClaw, Claude Code, MCP servers, custom Python tooling, RAG pipelines, Ollama optimization. If you need private inference (legal, medical, financial, defense-adjacent) without sending tokens to Anthropic or OpenAI, I have the stack running. Legacy backend rewrites and migrations. Fixed-price after a paid Discovery Sprint, never exploratory. I read .NET, ASP.NET Identity, SQL Server, and write Python. Drop-in replacements preserving schema and session model, parity contract tests, zero-downtime cutover. Scope locked at Discovery Sprint end. I also do smart contract security audits (Code4rena and CodeHawks track record, findings across 8 protocols in a single campaign week), low-level binary RE (PS2 C++ static recompilation, Halogen project), and anti-detection scraping at scale. I scope upfront, ship what was scoped, and build reliable, antifragile systems. Available when you need more.