Overview

In this role you will advance an autonomous research agent by designing an evaluation suite and data-to-training pipeline. You’ll post-train small open-weight models to improve generation, ranking and execution of scientific ideas, collaborating with the Founding Scientist and a focused engineering team in Barcelona. You’ll iterate on evolving data, publish results in journal clubs, and shape the Next-Gen roadmap to unlock capabilities in the Seqera Platform.

Compensaciones / Beneficios

- Versátil working hours
- Private health insurance
- Private life insurance
- Equity
- Home office allowance (valued over 1000 USD)
- Learning and development budget (1000 USD)

Responsabilidades
- Design and implement the evaluation suite to define “better” for the autonomous research agent
- Build a pipeline to convert deployment run data into a training-ready corpus
- Post-train small open-weight models (SFT, preference tuning, adapters) to beat baselines with lower cost and latency
- Run repeated training cycles on evolving data
- Participate in journal clubs and stay at the frontier of post-training specialized models
- Own the Next-Gen roadmap with the Founding Scientist,



defining data-to-capability sequencing

Requisitos principales
- Post-trained open-weight language models with SFT and at least one preference-tuning method
- Experience retraining models on changing data, handling regression and forgetting (replay, data mixing, adapter strategies, eval gating)
- Built evaluation harnesses for specific tasks and defended results
- Fluent Python and modern stack (PyTorch, transformers, TRL, PEFT, vLLM or equivalents)
- Strong open-source footprint or contributions to the mentioned tools, or fine-tuned models on the Hub
- Proactive operator mindset; ability to build pipelines when tooling is incomplete
- Willingness to work in-office in Barcelona or relocate

- Problem-solving and initiative
- Collaborative mindset and ability to defend results to skeptics
- Adaptability in data/tooling directions and ambiguity
- Post-training of open-weight LMs (SFT, preference tuning)
- Data replay and continual learning techniques (data mixing, adapters, eval gating)
- Evaluation harness design and metric definition

📌 Machine Learning Research Engineer (Barcelona)
🏢 Seqera
📍 Barcelona