About Luxima

Luxima is the Digital Product Passport platform built by MBP Systems, a deep-tech spin-out from University College Dublin. The EU's Ecodesign for Sustainable Products Regulation will require millions of products to carry a digital passport, and the small and mid-sized businesses it reaches first don't have compliance departments to cope. Luxima takes the product data a business already has — however scattered, inconsistent or half-missing — and turns it into compliant passports automatically. The "automatically" is the AI: it's the heart of the product, and it's what you'd own.

The role

You'll own the intelligence in the platform: the pipelines that read a customer's spreadsheets, datasheets and PDFs and turn them into structured, verifiable product facts — and just as importantly, the evaluation infrastructure that proves those facts are right. This is applied AI engineering with an unusually high bar for correctness: our output goes into legal compliance documents, so "usually right" isn't good enough and every claim needs a traceable source. You'll work directly with the founding team, including the researchers behind the underlying approach, from our base at NovaUCD.

What you'll build

  • Extraction pipelines — LLM-powered systems that pull structured product data out of spreadsheets, ERP exports, supplier PDFs and datasheets, with confidence scores and source attribution on every field.
  • Evaluation infrastructure — ground-truth datasets, regression suites and accuracy metrics, so we know precisely how good extraction is and catch it the moment a change makes it worse.
  • Structured generation — mapping extracted facts onto per-category passport requirements, working hand in hand with the rules engine.
  • Human-in-the-loop review — the logic that decides what the system is sure about, what gets flagged to a human, and how reviewer corrections feed back into quality.
  • Model strategy — choosing and combining models across providers, and tuning the cost, latency and accuracy trade-offs as volume grows.

What you'll bring

  • Hands-on experience shipping LLM-based systems to production — not just prototypes: systems with real users, real failure modes and real cost constraints.
  • Strong Python, and enough comfort with TypeScript to work inside our services when a pipeline meets the platform.
  • Fluency with structured output techniques — JSON schema, function calling, constrained decoding — and the prompt-engineering judgement to know when prompting stops being the answer.
  • An evaluation mindset: you don't believe a pipeline works until you've measured it, and you build the harness before you tune the prompt.
  • Solid data-wrangling fundamentals — messy tabular data, document parsing, entity matching.
  • Healthy scepticism about model output. In our domain, a confident hallucination is worse than a gap.

Nice to have

  • A background in classic NLP or information extraction — NER, entity resolution, document layout analysis.
  • Document processing at scale: OCR, PDF parsing, table extraction.
  • Fine-tuning or distillation experience, and knowing when it beats prompting a frontier model.
  • Retrieval systems — embeddings, vector search, RAG — in production.
  • A research background or publications in ML/NLP; we came out of a lab and it shows.

The engagement

This is a 6-month contract with a planned extension path to 12 months, tied to our platform roadmap. You'll contract through your own company or an umbrella arrangement (Ireland, UK or EU). We're based at NovaUCD, University College Dublin's innovation hub, and work hybrid — time in the office with the team, remote when it suits the work. The day rate depends on experience — tell us your expectations and we'll be straight about fit.

How to apply

Email info@mbpsystems.com with the subject line "Application: AI Engineer". Include a CV or LinkedIn profile, a link to something you've built with LLMs (a system, a repo, a write-up — anything real), and a few lines on the extraction or evaluation problem you're proudest of solving. No cover letter required. Our process is four steps and usually done inside two weeks — here's how it works.