←All roles

Senior AI Engineer — Models & Evaluation

You'll join as our Founding AI Engineer to take a working system and make it excellent

ABOUT THE ROLE

Bardin is the always-on application engineer for the robotics and automation industry. The knowledge graph, agents and data pipeline already run against real customer data. Our engineers build the agents and the orchestration around them in Mastra, and they do it well.

We need the person who makes the parts underneath those agents measurably better. That covers the models, the embeddings, the graph, and the evals that tell us whether any of it worked. This is a hands-on senior role for someone who has trained models themselves, not only called them through an API.

WHERE YOU SIT

You're the link between research and product. Our AI researcher explores what's possible. Our application engineer brings the automation-domain knowledge. Our full-stack team ships the agents. You take an idea from experiment to a measured, reliable part of the product, and you decide when the numbers are good enough to ship.

IN THIS ROLE YOU'LL

  • Own evaluation in depth. Build benchmarks and golden sets from real customer questions, write metrics that track what users actually experience, and report results with error bars. When a number moves, you can say why.
  • Fine-tune and distil small models. Build specialised models for classification, extraction, reranking and query routing that beat a large general model on our tasks, at a fraction of the cost and latency.
  • Make retrieval strong. Fine-tune embeddings on our domain (hard negatives, contrastive training), choose and tune rerankers, and prove each gain on a held-out set.
  • Take the knowledge graph from working to excellent. The team builds and extends the graph. You make it right: measure its quality, fix entity resolution and duplicates, sharpen the schema, raise extraction accuracy from technical PDFs and catalogues, and add checks that catch bad data before an agent reasons over it.
  • Benchmark before we commit. Compare models, providers, embedding models and graph approaches on our own data, and write up the results so the team can act on them.
  • Raise the bar for the team. Review the evals other engineers write in Mastra and turn research results into tools they can use. Python where the models live, TypeScript where the product lives; you're at home in both.

This is not the role for you if you mainly want to build agent workflows or product features. The team already does that well.

Requirements

  • 5+ years in ML or NLP, with real hands-on model work: you've trained, fine-tuned or distilled models yourself (SFT, LoRA, distillation; RL a plus).
  • You've designed evaluation sets and benchmarks from scratch, and you're rigorous about statistical significance, leakage and overfitting to the test set.
  • Deep experience with retrieval: embedding models, rerankers, hybrid search, and fine-tuning them for a domain.
  • Strong Python and PyTorch or JAX for training, evaluation and serving (GPUs or TPUs, vLLM or similar).
  • Comfortable in TypeScript: our product and agents are built on it with Mastra. You'll ship your models, evals and retrieval changes into that codebase, not hand them over in a notebook.
  • Experience making a knowledge graph or structured dataset trustworthy at scale: entity resolution, extraction quality, schema design, and measuring all three. Graph databases (FalkorDB, Neo4j, Cypher) are a plus.
  • You can explain a modelling result to non-specialists and defend it.

Nice to have: a PhD or published research in NLP, retrieval or LLMs; pre-training experience; industrial or B2B domain data.

Apply

Senior AI Engineer — Models & Evaluation

Send your CV and we'll take it from there. We read every application.

CV

Up to 10 MB.