Our client is looking for a Research Engineer to join their team.
Duration - 12 months
Location - Hybrid (3 days onsite)
Summary:
AIRA is a FAIR team building AI agents that speed up AI research itself. These agents can form hypotheses, write and run experiments, analyze results and iterate. We work on agent design, environments and benchmarks for research tasks, evaluation methods (including LLM judges and rubrics), and the infrastructure to run agents reliably at scale. We publish our work and also contribute to Meta’s wider agentic and coding efforts.
The role Research Engineers on AIRA work closely with Research Scientists to turn ideas into working, reproducible systems. As an IC4, you will own well-defined parts of the research stack from start to finish. That could be a benchmark, an agent component, an eval pipeline or a training or inference workflow. You will be expected to deliver high-quality, scalable code with growing independence.
Responsibilities
Design, build and maintain agent scaffolds, tools and environments for research tasks, such as ML engineering, coding and experiment execution.
Build and run large-scale experiments: launching jobs, tracking them and analyzing results across many GPUs.
Develop evaluation pipelines and benchmarks, including automated judges, rubrics and reproducibility checks.
Implement and extend training and fine-tuning workflows for LLM-based agents (SFT, RL and post-training).
Improve the reliability, throughput and cost-efficiency of agent rollouts and cluster usage.
Turn research prototypes into reusable, well-tested libraries for the wider team.
Contribute to papers, open-source releases and internal tech talks.
Take part in team oncall rotations, such as reproducibility and eval infrastructure.
Requirements
BS or MS in Computer Science, Engineering, Math or a related field, or equivalent practical experience.
2+ years of software engineering or research engineering experience, or a PhD with strong engineering output.
Strong Python skills and experience with PyTorch or a similar deep learning framework.
Hands-on experience training, fine-tuning or evaluating large language models.
Experience running distributed or large-scale jobs on GPU clusters.
Good software practices: testing, code review, version control and readable, modular design.
Preferred qualifications
Experience building LLM agents: tool use, planning, multi-step reasoning or code execution sandboxes.
Familiarity with agent benchmarks for coding or ML research, such as SWE-style or MLE-style tasks.
Experience with RL for LLMs or other post-training methods.
Experience designing evaluations, including LLM-as-a-judge, rubric design and statistical rigor.
Publications at top venues (NeurIPS, ICML, ICLR, ACL), or notable open-source contributions.
Experience working with academic or cross-functional partners.
Skills Technical
Python (expert), plus bash and Linux tooling
PyTorch, plus distributed training (FSDP, DDP) and inference serving (e.g., vLLM)
LLM fine-tuning, RL and post-training pipelines
Agent frameworks, tool and function calling, prompt and scaffold design
Sandboxed code execution, containers and job schedulers (e.g., SLURM)
Experiment tracking, data pipelines and large-scale result analysis
Benchmark and evaluation design, metrics and statistical testing
Profiling and optimizing GPU usage and throughput
Reproducibility practices: seeding, config management and artifact versioning
Research
Reading papers and implementing them quickly and faithfully
Designing ablations and clean experimental comparisons
Interpreting noisy results and turning them into next steps
Technical writing for papers, docs and reports
Collaboration and execution (IC4)
Owning medium-sized projects from start to finish with little supervision
Breaking down vague research goals into concrete engineering tasks
Communicating clearly with scientists, engineers and partners
Mentoring interns and newer engineers through code review and pairing
Giving honest estimates and flagging risks early
Benefits
Competitive salary
Healthcare contribution and inclusion in company pension scheme
Work laptop and phone
25 days annual leave (pro-rata) plus paid bank holidays
Expanding workforce with potential for career progression for top performers