Skip to content
View FullFran's full-sized avatar
🐧
Focusing
🐧
Focusing

Highlights

  • Pro

Organizations

@blakiatech

Block or report FullFran

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
FullFran/README.md

Francisco Olmedo — Physicist and AI / backend engineer

Francisco Olmedo

Physicist. I build backend and AI systems that have to work outside a demo.

Physics taught me to ask what is actually being measured. I bring the same habit to software: make the system understandable, test the important parts, and find out where it breaks before calling it done.

Currently
— Lead AI Engineer @ Hagalink — backend, agents and applied AI
— PhD candidate in applied AI @ Universidad de Córdoba · CIEMAT
— Working mainly with Python, TypeScript, Go, PostgreSQL, queues and RAG

Some things I've built
feedpulse — a queue-based ingestion service for more than 10,000 RSS sources
Hybrid-RAG-example — semantic and keyword retrieval fused with RRF
llm-pii-redactor — keeping personal data behind the application's boundary before it reaches an AI provider
agent-eval-tui — local evaluation and OTLP traces for agent runs

What I use
Linux (Pop!_OS), Ghostty, tmux, Neovim and zsh. I keep the setup reproducible with Nix and home-manager because rebuilding a machine should not be a ritual.
I also type on a 34-key split, which is either a useful tool or a very elaborate way of avoiding a normal keyboard.
fulfran-dots

What I care about
Architecture that survives contact with reality. Systems I can still reason about at 3am. Less magic, more engineering.

Elsewhere
CV → fullfran.com/cv · email

🇪🇸 En español

Físico. Construyo sistemas de backend e IA que tienen que funcionar fuera de una demo.

La física me enseñó a preguntar qué se está midiendo de verdad. En software intento hacer lo mismo: que el sistema se pueda entender, que lo importante se pueda probar y que sepamos dónde falla antes de darlo por terminado.

Ahora mismo
— Lead AI Engineer @ Hagalink — backend, agentes e IA aplicada
— Doctorando en IA aplicada @ Universidad de Córdoba · CIEMAT
— Trabajo sobre todo con Python, TypeScript, Go, PostgreSQL, colas y RAG

Algunas cosas que he construido
feedpulse — servicio de ingesta con colas para más de 10.000 fuentes RSS
Hybrid-RAG-example — recuperación semántica y por palabras clave fusionada con RRF
llm-pii-redactor — datos personales detrás de la frontera de la aplicación antes de llegar al proveedor de IA
agent-eval-tui — evaluación local y trazas OTLP para ejecuciones de agentes

Lo que uso
Linux (Pop!_OS), Ghostty, tmux, Neovim y zsh. Mantengo el entorno reproducible con Nix y home-manager porque reconstruir una máquina no debería ser un ritual.
También escribo en un split de 34 teclas, que puede ser una herramienta útil o una forma muy elaborada de evitar un teclado normal.
fulfran-dots

Lo que me importa
Arquitectura que sobrevive al contacto con la realidad. Sistemas que todavía puedo razonar a las 3am. Menos magia, más ingeniería.

En otros sitios
CV → fullfran.com/cv · email

Pinned Loading

  1. Hybrid-RAG-example Hybrid-RAG-example Public

    Teaching resource I built to train my students in production-grade Hybrid RAG — semantic + keyword retrieval with RRF, Clean Architecture.

    Python 2 1

  2. llm-pii-redactor llm-pii-redactor Public

    Reference implementation of PII redaction at the AI-provider boundary — stable pseudonyms, canonicalisation, NER port, fail-closed residual check and rehydration. TypeScript, no frameworks.

    TypeScript 1

  3. snow-mcrt snow-mcrt Public

    Monte Carlo photon transport for light in snow. Reproduces the canonical snow-optics benchmarks from first principles, validated against closed-form radiative transfer.

    Python 1

  4. feedpulse feedpulse Public

    RSS monitoring platform with hexagonal/DDD architecture — three runtimes (api/scheduler/worker), observability, queues, benchmarks and production-grade tests.

    TypeScript 1

  5. Agent-skills-POC Agent-skills-POC Public

    Built to upskill my team on agentic AI — framework-agnostic Agent Skills with Progressive Disclosure, MCP, and a stateless reasoning loop. Clean Architecture.

    Python 1

  6. claudeops-tui claudeops-tui Public

    Local TUI to track Claude Code usage, costs, and tasks (Go + Bubbletea + SQLite)

    Go 2