AI engineer

Put retrieval, agents and structured output inside client products, and prove they still work after the next change.

What you'll do

Design retrieval and agent features that read and write the client's Django or FastAPI models.

Choose the model, the prompt and the tools, and keep an evaluation set running in CI.

Use pgvector and the database the product already has, rather than a side system nobody will operate.

Measure quality, cost and latency, and say when ordinary search would do.

Ship in the same repository, review and demo as the Python engineers on the engagement.

You write production Python, not only notebooks.

You have shipped retrieval or an agent that people depend on, and you can show how you evaluated it.

You know when embeddings and a vector index help, and when they do not.

You can read a Django or FastAPI codebase and add the feature without a rewrite.

You have the right to work in the UK.

LangGraph, or a similar library, on a product that is already live.

pgvector, or another Postgres-native vector store.

A budget you have had to keep, where a cheaper model was the right answer.

An evaluation harness you can walk through.