Position Overview:

We're looking for a Junior AI Engineer to join our research team and work hands-on across the full lifecycle of large language models — from training and fine-tuning, to inference and deployment, to the applications and tooling built on top of them. This is a builder's role with a researcher's mindset. You'll ship real systems, but you'll do it the way good science is done: with hypotheses, measurements, and a healthy skepticism toward numbers that look too good. It's a strong fit for someone early in their career who has solid fundamentals, a portfolio of real work (projects, research, internships, or open-source), and the appetite to grow quickly alongside an experienced team.

• Build RAG and agentic applications. Design, build, and iterate on retrieval augmented generation and agentic systems in healthcare.

• Optimize inference and own deployments. Improve LLM inference along latency, throughput, and cost — reasoning carefully about memory, batching, and data movement — and partner with the engineering team to manage model serving, deployment, and versioning end to end.

• Help make our training frameworks faster. Profile our training stack — including our reinforcement learning framework — to understand where the time and memory actually go (compute, communication, data movement), and contribute performance improvements alongside the team.

• Run distributed LLM training. Execute large-scale, multi-node training jobs across continual pretraining, SFT, and RL, following our datasets and recipes, and surfacing issues in the training dynamics as they appear.

• Build the surrounding stack. Develop the software and tooling that turns a model into a complete solution — retrieval pipelines, tool use and function calling, evaluation harnesses, orchestration, and the rest of the scaffolding that gives an LLM its real-world capabilities. 1 How we work This is the part we care about most, and it's what separates a good fit from a great one.

• Scientific approach. We treat problems as experiments. We form a hypothesis, design a clean way to test it, and let the results — not our assumptions — decide what's next.

• Rigor. We measure before we conclude. We're careful about evaluation, wary of contamination and overfit benchmarks, and we care whether a result actually generalizes rather than just whether it looks impressive.

• Critical thinking. We read papers and results closely and skeptically, we ground claims in evidence, and we're comfortable saying "I'm not sure that's true — let's check." Being right matters more than being confident.

• Teamwork. Research and engineering work as one team here. You'll collaborate across both, share work early, give and receive critique generously, and document clearly enough that others can build on what you do.