Location: Gipuzkoa, Spain (CET)

Remote: Yes, remote only (EU / UK / US-overlap hours)

Willing to relocate: No

Technologies: Python, FastAPI, PostgreSQL/pgvector, SQLAlchemy, Alembic, Celery, Redis, LangGraph, MCP, RAG (hybrid BM25 + dense, reranking), Claude/OpenAI APIs, Docker, GitHub Actions, TypeScript, Next.js. Learning Rust.

Résumé/CV: https://www.emanuelecz.dev/Emanuele-Czofei-CV.pdf

Email: emanuelecz06@gmail.com

Self-taught backend engineer, building since 2024. I build AI features into working software: retrieval, agents that call real tools, and the Python/Postgres services underneath. I speak Italian, Romanian, Spanish and English.

Shipped and paid for: servibai.com, a Next.js site for a home-repair business in Gipuzkoa. I built and launched it alone, and it has brought in about 5 new customers so far. Before that I did a remote internship at Gaddr, a Stockholm AI startup, where I built auth pages and Zod validation for an external API integration.

Docent (prototype) is a support agent for pydantic/pydantic. It drafts replies to new issues using ~1,500 issues that were actually solved. It uses pgvector plus Postgres full-text search, fused and reranked, and a LangGraph flow (intake → research → draft) whose research step calls the GitHub MCP server through a read-only allow-list. The part I'd talk about: the drafts were fluent but cited the wrong issues. The prompt wasn't the problem. I was storing closed tickets instead of solutions. Now it stores the actual fix (closing-PR body, a filtered last comment, or nothing) and embeds only the original question. https://github.com/emanuelecz/Docent-Backend

Also: RAG over the Spanish INSHT/NTP workplace-safety corpus. On a 50-question set I labeled by hand it scores Hit@8 0.98 and MRR 0.955, and the retrieval checks run in CI on every PR. https://github.com/emanuelecz/rag_project

Looking for: a hands-on engineering role (junior or contract) on a small team: backend, AI features, devtools. I want to be writing code. EU citizen, EOR or contractor.