I'm Luis Adrian, a self-taught backend engineer in Buenos Aires building my own agentic stack: context transport, retrieval substrates, and the operational glue that keeps multi-agent systems honest in production.
Ten years operating in high-consequence environments before I touched a keyboard. It shows in how I design — failure modes first, happy paths last.
Self-taught. Linux-native. CLI-first.
All open source. Each one is a deliberate study in a different layer of the stack.
Self-BOT and Meridian both live on a single GCP VM. The cards are a portfolio snapshot of the runtime shape, uptime window, and operating profile.
Short technical write-ups built from real operating questions: tail latency, retrieval quality, and provider routing.
Six weeks tracking a single tail-latency regression across the merge stage. The culprit was a synchronous protobuf re-serialization sneaking in on fan-out.
Combining vector RAG with recursive CTE graph traversal on a 12k-doc corpus. The graph hop fills what cosine similarity misses.
Provider routing log over March 2026. Local Ollama handles half; the heavy reasoning still goes to Claude.
The portfolio stays readable as a single overview. Larger interactive pieces live on dedicated endpoints where backend integrations can grow without crowding the homepage.