About the role
You build systems that answer questions from a company’s documents and carry out agreed actions in its systems. You start with a data audit rather than with the choice of model, because the outcome is decided by the state of the documentation, not by the name of the model.
You are responsible for the three things that most often sink projects of this kind: the cost of handling one case at full traffic, how the system behaves when it is unsure of an answer, and whether the client’s team can take the project over.
What you will be doing
- Assessing whether the client’s documentation can be searched by machine, and putting it in order when it cannot
- Building document search and assistants that show the source of every answer
- Setting the confidence threshold and the path for passing a case to an employee
- Agents with write permissions in the client’s systems, together with the scope of those permissions and a record of what they did
- Measuring unit cost and answer quality before the decision to go live
- Logging events in a way that makes it possible to reconstruct where a given answer came from
- Choosing a processing model that meets the client’s requirements, up to and including a model run on their own infrastructure
What we expect
- Practical experience with language models in projects delivered to clients, not only in experiments
- Python at a level that lets you build and maintain a production service
- Familiarity with full-text and vector search, and the judgement to choose between them
- The ability to measure answer quality on a test set instead of judging it by feel
- Awareness of the limits: hallucinations, context cost, latency, data protection
- Polish at C2, English at C1
Nice to have
- Experience running models on a client’s own infrastructure
- Knowledge of GDPR requirements for processing company documents
- The ability to explain a technical decision to a non-technical person
How recruitment works
An introductory call, a task on an anonymised set of documents, and a technical conversation. We answer every application within 7 working days.