AI Engineer & CTO — LLM Agents, RAG & Production GenAI | LangGraph, PyTorch | PhD Researcher

I build generative AI systems that reach production — not demos.

WHAT I DO

• LLM agents and multi-agent systems — LangGraph and LangChain architectures with planning, tool use and memory, split by failure mode so each stage can be evaluated and debugged independently.

• RAG over regulated document sets — mandatory citation, verification of the answer against the retrieved chunk, and an explicit "I don't have this figure" instead of a plausible invention.

• LLM evaluation and monitoring — benchmarks, hallucination detection, chain-of-thought analysis. This is also my doctoral research, so it is designed into the systems I ship rather than bolted on afterwards.

• Generative AI for biotech — diffusion models, graph neural networks and protein language models.

• Process automation with generative AI — document extraction, structured generation, end-to-end workflows.

CURRENT WORK

CTO at an AI startup, where I build the agent runtime, data pipelines and validation layers behind a document-automation platform in a regulated domain. Measured in pilots: 70% less time per reporting cycle, 3x higher response rate on automated outreach, and 100% of reported figures traceable to a source document. Where an auditor reads the output, a plausible-but-wrong answer is worse than no answer, so grounding and traceability were design constraints, not features.

In parallel, AI Research Engineer and PhD researcher at INAB Barcelona, working on LLM internal reasoning and hallucination detection.

BEFORE THAT

Co-founder and CTO of ProtForge (2023-2025), a generative AI platform for antibody design: diffusion models, GNNs and protein language models to generate and rank candidates. The engineering went into ranking rather than generation — sampling the sequence space is tractable, prioritising it well is the real bottleneck. MBA at Lanzadera in parallel.

MSc Computational Cognitive Science (University of Groningen), BSc Biomedical Engineering (UPV). Four first-place finishes in AI hackathons and startup competitions, most recently HackBarna 2025.

HOW I WORK

Remote-first from Barcelona (CET). B2B invoicing through my own Spanish company — intra-EU reverse charge, no umbrella company and no payroll overhead on your side. Available part-time immediately, full-time for long-term engagements, and able to travel within Europe when a project needs it.

STACK

Python, LangGraph, LangChain, AutoGen, CrewAI, MCP, RAG, vector databases, PyTorch, transformers, diffusion models, graph neural networks, NLP, fine-tuning (SFT, RLHF), prompt engineering, LLM evaluation, MLOps, AWS, GCP, Docker, CI/CD, PostgreSQL, REST APIs. TypeScript and Next.js when a system needs a front end.