You'll take models from notebook to production. That means LLM-powered features, retrieval pipelines, classical ML where it's the right tool, and the evaluation discipline to know which is which. We care about shipped systems, not leaderboard scores.
What you'll do
- Design and build RAG, agentic and fine-tuned LLM applications end to end
- Train, evaluate and iterate on classical ML and deep-learning models
- Build data pipelines for feature engineering, labelling and evaluation sets
- Deploy models as services — containerised, monitored, versioned
- Set up offline + online evaluation: metrics, regression suites, human review loops
- Optimise inference cost and latency; pick the right model for the job
Must have
- Strong Python and at least one of PyTorch / TensorFlow / scikit-learn
- Solid ML foundations: bias–variance, regularisation, evaluation metrics, data leakage
- Hands-on with LLM APIs or open-weight models (prompting, embeddings, RAG)
- Comfort with Git, Docker and a cloud provider (AWS / GCP / Azure)
- A portfolio, GitHub or project you can walk us through in depth
Nice to have
- Vector databases (pgvector, Pinecone, Qdrant) and LangChain / LlamaIndex
- MLOps tooling: MLflow, Weights & Biases, feature stores
- NLP or computer-vision specialisation; published work or Kaggle rank