The Role :

Work at the forefront of medical AI - not by wiring together off-the-shelf services, but by owning the systems and the models behind them. Clinical work is messy, spoken, time-critical, and unforgiving of errors, which makes it one of the most interesting applied AI problem spaces there is.

The headline opportunity: pushing the frontier on special-purpose models of our own - fine-tuned, highthroughput, domain-adapted models that beat general-purpose LLMs on the narrow things medicine actually needs, at a fraction of the cost and latency.

Prior fine-tuning experience is not required; we want strong applied AI engineers with solid fundamentals and the motivation to grow into this, and we'll support you in doing so.

What you'll do:

- Real-time clinical document generation - turn a live consultation into a structured, clinically accurate document while the clinician is still in the room. Sub-second latency, zero tolerance for hallucination.

- Fine-tuning special-purpose models - LoRA, PEFT, SFT and preference alignment, optimized for highthroughput serving. Dataset design through to production. This is where we invest in going deep.

- Speech-to-text and audio pipelines - real clinical audio: multiple speakers, accents, noise, and Swedish and German medical vocabulary no off-the-shelf model has seen.

- Medical NER and structured extraction - medications, dosages, diagnoses, procedures and temporal relationships pulled out of free text, reliable enough to write into a patient record.

- Evaluation pipelines - in healthcare, "it looks good" is not a metric. Golden datasets, automated judges, clinician-in-the-loop review, and the monitoring that catches regressions before a clinic does.

- Automated booking and scheduling - agentic systems that handle real patient interactions end to end, with real-world consequences when they go wrong.

- Data and deployment - curate clinical datasets to a training-ready standard, and take systems from experiment to monitored, production-grade services clinicians rely on at scale. You won't do all of this at once, but over time you'll touch most of it.

What we're looking for :

- Bachelor's in Computer Science, Engineering, or a related field.

- 3+ years in AI/ML, with a focus on applied and Generative AI systems.

- Strong hands-on experience with LLMs, prompt engineering, and structured outputs.

- Hands-on with speech-to-text and audio-processing pipelines.

- Proficient in Python (pandas, NumPy, ML/AI libraries) and SQL.

- Familiar with cloud AI services, preferably AWS (S3, SageMaker, Bedrock).

- Understanding of LLMOps / MLOps: evaluation, monitoring, deployment, and iteration.

- Proven experience shipping production-grade AI systems at scale.

- Strong ownership, clear communication, and independence in a fast-paced environment.

Good to have :

- Parameter-efficient fine-tuning (LoRA, PEFT, SFT) and tools such as Hugging Face Transformers, PEFT, TRL, Axolotl, or Unsloth.

- RLHF, DPO, or other preference-alignment methods.

- Information extraction / NER, and experience with clinical or biomedical ontologies.

- End-to-end GenAI workflows (LangChain, LlamaIndex, RAG), embeddings, vector databases, and retrieval.

- Agentic systems, tool use, and function calling in production.

- Model quantization, distillation, or inference optimization; distributed / multi-GPU training.

- Real-time / streaming inference and low-latency serving.

- Healthcare, clinical, or other regulated-data experience.