Skip to content

About me

The short version

I help teams take AI systems from "it works in a notebook" to "it runs in production and we trust it."

If your prototype already proves the value but the path to a reliable, maintainable system still feels unclear - that is exactly where I work best.

Portrait of Andrés Espinosa Sanfiel, AI Engineer

How I got here

I studied Economics before I ever touched a line of Python. At the time, it felt unrelated. It was not. Economics trained me to think in trade-offs, incentives, and measurable outcomes - which is exactly what matters when a business is about to bet real money on an AI system.

I moved into data engineering first, building Spark pipelines and learning what it actually takes to keep data flowing reliably at scale. Then data science - ML models, cloud AI services, MLOps. And eventually AI engineering, where I found the work that makes me want to keep solving problems: taking AI that impresses in a demo and turning it into something people can trust every day.

Here is what that path taught me. The model is rarely the hard part. What breaks projects is everything around it - the routing, the retries, the system boundaries nobody defined, the retrieval that works perfectly on test data and fails on real queries. I have been on both sides of that problem enough times to know where the risks hide.

I hold a Master's in Advanced and Applied Artificial Intelligence from the University of Valencia, and a Bachelor's in Economics. One gives me the technical depth, the other gives me the habit of asking "does this actually make business sense?" before writing a single line of code. I use both daily.

What I have been building

Right now, my work sits in the gap between a working AI prototype and a system a team can actually depend on.

Recent projects include agentic workflows that cut manual processing by 70%, hybrid RAG pipelines that improved domain-specific retrieval accuracy by 40%, and FastAPI services handling 500+ daily API requests for live AI workloads.

One of those - a production email automation system built with PydanticAI, RAG, and FastAPI - I presented publicly at the Datamecum Webinar 2025. It was my first technical talk in front of an audience, and honestly, it was a turning point. Speaking publicly about what I build forced me to be clearer about why each design decision mattered - and that clarity now shows up in every project.

You can explore the full details in my case studies.

Certifications & Speaking

Core Stack

  • Agentic AI & Orchestration

    LangChain, LangGraph, PydanticAI, Agno, MCP protocol, deterministic workflows, multi-step agents, and tool orchestration.

  • Retrieval & Data

    OpenSearch, Qdrant, PostgreSQL, pgVector, PySpark, SQL, and domain-specific search architectures that combine vector and keyword retrieval.

  • Platforms & Cloud

    Python, FastAPI, WebSockets, AWS (ECS, Lambda, S3, DynamoDB, EventBridge), Azure, Docker, Redis, and backend APIs built for real usage.

  • Delivery & MLOps

    CI/CD, MLflow, Langfuse, LangSmith, model tracking, deployment workflows, and hexagonal architecture for maintainability and decoupling.

How I work

Every project starts with a simple question: what does success look like, and how will we measure it? The architecture follows from the answer - not the other way around.

I am opinionated about a few things. I prefer deterministic workflows over black-box magic. If the system behavior cannot be tested and observed, it is not ready for production. That means I think carefully about where the LLM adds real value and where plain logic does the job better - and I am not afraid to say "you don't need AI for that part."

I build for production from day one. Observability, maintainability, and clear system boundaries are not things I bolt on at the end - they shape the initial design. When the engagement ends, the goal is always a system your team can run, extend, and evaluate without needing me in the room.

I work well on my own and also embedded with an internal engineering or product team. Either way, I will tell you what I think will work, what will not, and what the trade-offs are. I would rather have an honest conversation now than a painful one three months into delivery.

Languages & Availability

  • I work in English and Spanish.
  • Based in Spain and available for remote collaboration across Europe.
  • Open to freelance, contract, and selected long-term engagements.

Want to discuss a project?

If you already have the business case but the delivery path still feels fragile, let's talk. I can help evaluate the technical shape, the risks, and the next steps before your team commits to implementation.

Book an intro call Connect on LinkedIn