Founding AI Engineer

Talent R; Barcelona; On-site; EUR 70,000 - 120,000; Full time

Job summary

Talent R is a European AI startup seeking a deep AI engineer to own AI flows across the product, from document extraction to generation. You will explore state-of-the-art models, evaluate architectures, and ship production-ready improvements that influence customers' high-stakes decisions.

Ideal candidates have hands-on experience with retrieval, extraction pipelines, and multi-agent systems, plus a track record of production AI work and rigorous evaluation.

Qualifications

• Deep AI expertise with focus on AI systems.
• Experience building retrieval/extraction pipelines and multi-agent systems.
• Proven production-grade AI deployments (>1 year).

Responsibilities

• Own AI flows across the product: document extraction, retrieval, agentic workflows, generation.
• Build the evaluation harness before scaling: golden datasets, regression tests, error taxonomies, LLM-as-judge where warranted.
• Own quality/latency/cost trade-offs across models and vendors.
• Fine-tune when it wins — decide when smaller specialized models beat frontier ones.
• Decide when not to use an LLM: replace probabilistic steps with deterministic logic.
• Track field developments and ship improvements.

Skills

Deep AI expertise

AI systems

Retrieval systems

Extraction pipelines

Multi-agent systems

Evaluation

Python

PyTorch

TensorFlow

Education

PhD or strong research background

Tools

Python

PyTorch

TensorFlow

Job description

European AI startup · Seed-funded · Deep AI specialist

The company

A fast-growing European AI startup building the platform that helps companies find, assess and win public tenders — public procurement is roughly 10% of EU GDP, and for most companies it's completely opaque.

Seed-funded, with product-market fit in their home market and real commercial traction. Now expanding across Europe. Around 30 people, 14 in engineering, both founders still write code. The product is agentic end to end.

The role

This is a deep AI role, not a broad engineering one. You own every AI flow in the product: extraction from messy real-world documents, semantic search, generation.

The bar is high because of what's at stake — customers make six-figure decisions on what the models extract. "It usually works" isn't a result. You make it accurate, fast and cheap, and the finish line is production, never a write-up.

What makes this seat specific: you're expected to explore the state of the art and bring it into the product. New models, new techniques, new architectures — you evaluate them, decide what's worth shipping, and ship it. That's the half of the job a general AI engineer doesn't do, and the half a pure researcher never finishes.

What you'll actually do

• Own the AI flows across the product: document extraction, retrieval, agentic workflows, generation
• Build the evaluation harness before scaling anything: golden datasets, regression tests, error taxonomies, LLM-as-judge where it's warranted
• Own the quality / latency / cost trade-off across models and vendors
• Fine-tune when it wins — you'll have the data, the budget and the freedom to decide when a smaller specialised model beats a frontier one
• Decide when not to use an LLM: replacing a probabilistic step with deterministic logic is often the right call, and knowing where that line sits is part of the craft
• Track what's happening in the field and turn it into shipped improvements

What we're looking for

• Deep AI expertise. Not a generalist who added LLMs to a backend role — someone whose core craft is AI systems
• You've built things that worked: a retrieval system, an extraction pipeline, a multi-agent system that survived real users
• You live in evals: no prompt or model change ships without knowing what it improved and what it broke
• You own the whole pipeline, not just the model call: parsing, retrieval, structured output, fallbacks, cost
• From a new model release to a working prototype in days
• You read papers and turn them into production decisions, not into write-ups
• At least 1 year of AI in production. Background matters less than depth — PhD, research experience or self-taught all welcome, provided the systems are real

What you won't find here

No micro-management — you own your stack. No 9-to-5 mentality — outcomes over hours. No "we've always done it this way."

Stack

Practical

• Barcelona preferred; Spain, UK or France considered