10/27/2026 updated
100 % available
Senior AI Engineer – LLM Agents, RAG & Evaluation (Python) – 100% remote
Mutilva, Spain
Worldwide
Telecommunications Engineer + Master's in Communication Research, UPNAAbout me
I build production AI systems: agents that call tools and APIs, retrieval over business documents (RAG), and the evaluation that proves they still work after every change.
AI AgentsRAGLLMLangChainPython (Programming Language)PostgreSQLSQL DatabasesGoogle CloudMulti-Agent SystemsFastAPIDocker
I build production AI systems: agents that call tools and APIs, retrieval over business documents (RAG), and the evaluation that proves they still work after every change.
For the past 3 years I have delivered applied-AI systems end to end for clients in energy, installations, legal services and manufacturing: a voice-to-order agent in production, a RAG assistant for field technicians over ~130 manuals, and schema-validated document extraction later reused as a tool in an email-to-offer workflow. I work evaluation-first: golden sets, confidence intervals and regression gates.
You can see how I evaluate in my public repo full-purchase-cycle-agents (github.com/mikelju/full-purchase-cycle-agents): a LangGraph purchase cycle (email/web/WhatsApp orders, supplier quotes, invoice reconciliation) with human approval, tested by 21 replayable eval suites on audited golden sets of up to 2,000 cases, with Wilson 95% confidence intervals, baseline gates and an exact McNemar regression test.
Before AI, I spent 8 years in wind-turbine data analysis and Python tooling at Siemens Gamesa and 8 years in engineering research.
Typical engagements:
- Taking an LLM prototype to production with evals, tracing and cost control
- Designing tool-using agents with human-in-the-loop approval
- Building or fixing RAG pipelines and measuring retrieval quality
Stack: Python, FastAPI, LangGraph, Pydantic, PostgreSQL/pgvector, LangSmith, OpenTelemetry, Docker, GCP, Azure.
Portfolio: github.com/mikelju
Experience
Professional experience10+ years
Freelancer since2026
Languages
EnglishFluentSpanishNative speaker
Project history
Purchase-cycle agents – public project
LangGraph agents for orders, supplier quotes and invoice reconciliation with human approval, plus 21 replayable eval suites with confidence intervals and regression gates.
LangGraph, Pydantic, LangSmith, OpenTelemetry
Voice-to-order agent – industrial contractor
Voice → LLM extraction of items and quantities → pgvector catalogue search → draft order approved by the user. In production.
Python, FastAPI, Whisper, pgvector, human-in-the-loop
Invoice extraction – energy consulting
Heterogeneous electricity bills turned into schema-validated JSON, later reused as a tool in an email-to-offer workflow. In production.
Multimodal LLM, JSON Schema, FastAPI, Supabase