Last month, haddock users consumed +20 billion tokens through our AI product. This isn't AI hype. It's our AI workforce, running 24/7 for thousands of restaurants across Spain.

None of it starts from clean data. Thousands of suppliers, no standards, documents that contradict each other, an entire industry's economics buried in paper and spreadsheets. Pulling fields out of a PDF is the easy part. The challenge is building an AI system that can interpret a chaotic document and knows when to ask instead of guess.

We're not adding a chatbot to an ERP. We're building the intelligence layer that makes an entire industry legible, for the first time.

We're looking for an AI Engineer who is analytical, quantitative and deeply AI-native. Someone who parallelizes hypotheses with coding agents, stays relentlessly focused on AI that scales and solves real customer pain, and measures before optimizing without ever letting analysis slow down shipping.

➡️ Who are we?

haddock is the AI-powered back-office for restaurants. We started as a system of record (inventories, purchases, invoices, bank reconciliation, P&L reports) and turned it into the harness where our AI workforce lives. They run 24/7, doing the heavy operational work so restaurateurs can focus on what they do best.

Our customers don't just save time. They make better decisions, catch problems before they escalate, and run tighter businesses: an owner finally sees what they really pay for each product, where the margin goes, and the exact moment a dish stops being profitable. That's the kind of value we're here to deliver.

👉 Our mission is to empower millions of restaurateurs to level up their business.

👉 Our vision is to lead the market through innovation, strong relationships, and by becoming the go-to tech partner for restaurant success.

👀 haddock is backed by Y Combinator, the accelerator behind Airbnb and Twitch. We're growing fast with support from JME, Decelera, Extension Fund, Wayra, and Zone2Boost.

➡️ Why are we hiring?

Our ambitions are growing: more customers, more AI automation, more product surface, more value. Our engineers don't wait for a project to be defined. They create momentum, propose ideas, and are a key part of the company's growth.

We're growing very fast, which means hundreds of processes, decisions, and workflows every week that can and should be automated. From simple deterministic pipelines to AI agents that get the job done. That's where you come in: an AI Engineer who loves solving complex challenges by designing, building, and deploying AI systems that think, act, and automate intelligently.

➡️ What you'd be working on

So you know what the job actually is:

• Document understanding at scale. Every invoice and delivery note a restaurant receives goes through our OCR. Accuracy at single field level, business casuistics for document understanding, cost per document and latency are all on the table, every single day.

• Fina, our conversational agent, and the most ambitious thing we build. Fina sits on top of everything haddock knows about a restaurant: purchases, suppliers, stock, sales, bank movements, P&L. Ask it anything about the business and it answers, and increasingly it goes and does the work instead of just reporting on it. It's the surface where the whole system has to behave as one, and most of what it will become is still unbuilt.

• End-to-end workflows that get the job done. Complex reconciliations, data workflows, and imports of the most chaotic data you can imagine: a supplier catalogue inside a photographed PDF, a POS export nobody ever documented, ten years of a restaurant's history in a spreadsheet built by six different people. Somebody has to turn all of that into something a system can trust.

• Product and format matching. Turning the messy free text of thousands of suppliers into a clean catalogue a restaurant can run its inventory on.

• The internal packages every one of our agents shares: AI observability, evals and human annotation. They are ours, we maintain them, and the quality of every agent we ship depends on them.

➡️ What you'll do

• Architect and ship AI solutions for restaurateurs, balancing latency, cost, and accuracy.

• Obsess over efficiency. A single line of code has saved us thousands of dollars, minutes of latency, or unlocked a step change in accuracy. You'll think this way by default, and you'll keep the cost of what we run in view as we scale.

• Own AI observability end to end: monitor billions of traces, design evaluation frameworks, and close the improvement loop as fast as possible.

• Maintain and develop our internal packages for AI evals, AI observability, human annotation, domain . Every agent we run is built on top of them, so making them better makes everything better at once.

• Build the harness. A model on its own is not a product. The tools it can call, the data it can read, the guardrails, the place where a human steps in: you build the backend, the frontend and the plumbing that turns a model into something thousands of restaurants rely on every day.

• Research and experiment. New models and techniques land every few weeks and some of them change what's possible for us. Running the experiment, reading the numbers and deciding whether it's worth it is part of the job, not something you get to when there's time left over. We'd rather find out first than read about it later.

• Talk to customers to validate new solutions and get feedback.

➡️ What you need to bring

👉 2+ years of hands-on experience with GenAI and 3+ years of engineering experience overall. One year of very intense AI work also counts, if it's backed by solid software or data science experience.

👉 You've run GenAI in production at real scale. As a reference: an agent or workflow doing at least 5,000 executions a month, and you were the one accountable for it. Experiments don't count.

👉 You've worked with an eval system to measure and improve model performance, rather than a handful of test prompts and a good feeling.

👉 You're truly AI-native: you follow new models, papers, and techniques, and you continuously refine how you use them in practice.

👉 You're also AI-native in how you build: you use modern AI-powered dev tools (Claude Code, Cursor, Codex or similar) and assemble your own agents and workflows to move faster.

👉 You combine engineering rigor with product thinking. You design, debug, and improve end-to-end systems that solve real problems.

👉 Fluent in Spanish and comfortable in English.

⭐️ You'll stand out if you have experience with

• AI observability platforms like langfuse (tracing, metrics, evals for LLM apps and agents)

• Designing and running evaluations

• Multimodal LLMs, especially vision models for OCR document understanding

• Building and orchestrating multi-agent systems

• RAG systems and vector databases

• Designing or working with MCPs

➡️ Our tech stack

TypeScript, Python, langfuse, PostHog, Google Cloud Platform, mastra, Postgres, MongoDB, and several AI providers.

➡️ What we offer

• 🚀 First and foremost, the project itself: a fast-growing startup with ambition, product-market fit, and great culture

• 🏢 A bright modern office in Barcelona, Poblenou, probably the best office you've ever worked at

• ☯️ At least 2 days per week at the office (in-person collaboration) + 3 days WFH for focus

• 🪄 All AI tooling you wish to experiment with

• 🧘 A subsidized Urban Sports Club membership: yoga, HIIT, hot studios

• 💰 Salary benefits via Playflow

➡️ Our Hiring Process

We move fast and communicate openly. Our goal is to wrap up the process within two weeks.

1. Intro call with Enric (AI Lead), 25 mins.

2. Call with Pol (co-founder, CPO), 25 mins.

3. Technical interviews with Enric and Guillermo (Head of Tech), 2 hours.

4. Final call with Arnau (co-founder, CEO), 25 mins.

Sounds like an opportunity? Apply now. We'd love to meet you.