Jordimp & Co. is the working name of one engineer in Barcelona. I build backend services and event pipelines, and more recently applied AI. Every AI feature I ship comes with the evals to check it. Java and Python are my production languages, TypeScript and Node.js turn up in modernization work, and my repos are public: CI, specs, receipts.
ggufbench: agentic benchmark for local GGUF models. Scores speed, tool-calling, code and multi-step agent tasks under one harness, with three runs per task and the raw receipts of every call. Live scoreboard at jordimarsal.github.io/ggufbench.
codebaserag: hexagonal RAG over a codebase. Retrieval quality is a CI gate: 40+ golden Q&A pairs and a committed recall@5 baseline that fails the build on regression.
harness-standard: multi-agent harness with spec-author, implementer and reviewer roles, and human gates between them. Installs into any project with one command.
kafka-adapter-telemetry: Kafka gateway and hub with idempotent Oracle persistence, dead-letter queues, health alerts and a live SSE dashboard. Java 25, Spring Boot 4.
mcp-transparent-png: MCP server in Python for PNG transparency, with three modes (rgb, auto, greenscreen).
redis-toolkit: rate limiting for LLM gateways. Atomic per-client token buckets that degrade gracefully when Redis fails. Java.
Site: jordimp.net · LinkedIn: jordi-marsal-poy · hello@jordimp.net


