AI / LLM Engineer
Take retrieval assistants and agents from prototype to production on clients’ governed data, evaluated before every release and traced once they run.
About the role
You build the AI side of DATAI’s client projects: assistants that answer from a company’s own documents, agents that call its internal tools, and the evaluation and tracing that show whether either is any good. You work next to the data engineers who prepare the sources, and a senior architect reviews your merge requests.
Projects and stack
The work starts from something that worked in a notebook and now has to work for real users. That might be a retrieval assistant over contracts or policies that cites its sources and respects each user’s permissions, or an agent that drafts an action and waits for a person to approve it.
The models come from OpenAI, Anthropic Claude, Amazon Bedrock, Azure AI Foundry or Vertex AI, depending on the client’s cloud. Orchestration is LangChain, LangGraph or MCP. Evaluation and tracing run on Langfuse, MLflow or Weights & Biases.
What you will do
- Build retrieval pipelines: document parsing, chunking, embeddings, hybrid search and reranking.
- Design agents and tool calls, with a person’s approval before any risky action.
- Build test sets with the client’s subject experts, and run them before every release against agreed quality scores.
- Add input and output guardrails, and tie access to the client’s identity system.
- Trace the latency, cost and quality of each request in production, and fix what the traces show.
- Demo to the client every week, and write the runbook for the team that will run the system.
What we look for
- Production experience with LLM applications: something real users relied on, not only a demo.
- Strong Python and the engineering around it: tests, typing, packaging and CI.
- Hands-on retrieval work with embeddings and vector or hybrid search, and an idea of where answers go wrong.
- The habit of measuring first: you set up an evaluation before you argue about a prompt.
- Working SQL, and an understanding of how data reaches your system.
- Clear written English for client tickets, reviews and documentation, and enough spoken English to explain a trade-off on a call.
Good to have
- Classical machine learning with MLflow or SageMaker.
- Work on one of the managed platforms: Bedrock, Azure AI Foundry or Vertex AI.
- Fine-tuning, or serving smaller open-weight models where cost or privacy calls for it.
- Consulting or client-facing work.
How we work
You join the client’s stand-ups and work in their repository and ticketing. You work in the US time zone, so stand-ups and reviews happen during the client’s business day.
No release goes out without passing its test set, and every request is traced in production from the first deployment. A senior architect reviews each merge request. Before handover you write the runbook and record the training, so the client’s team can run the system without us.
What we offer
You can work from Tashkent or remotely, on AI systems built for real users, and a senior architect reviews your code on every project.
Apply
Send your CV and, if you have them, links to your profile or code.
- 1You applyYour details and CV go only to the DATAI hiring team.
- 2We read itAn engineer reviews every application.
- 3First interviewA call with an engineer about your experience and the projects we staff. Compensation is discussed here.
- 4Technical interviewA working session with one of our architects: you design a retrieval assistant for a client scenario, say how you would evaluate it and what you would trace, then work through a short Python exercise.
- 5OfferIf both sides want to go ahead, we send a written offer.