AI Engineer

What you would own, end to end

You take a scoped AI requirement, design the system, build it, ship it to production, and maintain it. The full arc, not just a slice. For one business, that might mean an extraction pipeline that turns handwritten intake documents into structured records with gaps declared and never invented. For another, it might mean a retrieval system that answers compliance questions and cites the exact clause in the governing document. For another, it might mean a triage layer that reads incoming requests and routes them before a human ever touches them.

You own the evaluation harness that decides whether any of those outputs are good enough to show a user. You build the human-review paths for when they are not. If an extraction returns a wrong value, you are the one who built the safeguard that caught it, or the one who understands why it did not. That responsibility is the job.

The work and why it pulls

The AI work here is applied and operational: it lives inside systems that real businesses depend on every day. Extraction pipelines that pull structure from documents with no standard schema. Retrieval architectures that surface the right fragment from thousands of pages and attach the citation the user can verify. Evaluation harnesses that measure accuracy against a golden set of known-correct outputs before any user sees a result. Human-review flows that give a human the final call on results that need it, built in from the start.

The problem is not that the models are weak. Getting a model to be reliably useful inside a real operational workflow, across all the edge cases a business generates, is genuinely hard, and most of the industry has not solved it for businesses of this kind. Entoura has. The work now is holding that bar across more builds and more clients, and that bar is high. That is what makes the engineering matter.

In your first months, expect to build an extraction or retrieval feature from scoped requirement to production release, write and run the evaluation harness against a curated golden set of representative, de-identified inputs, and ship a human-review path that a non-technical user can actually operate. Where access to real client data is needed later in the process, it is least-privilege and controlled.

Who you would work with and where it leads

The studio is a senior team that stays deliberately lean. You would work alongside engineers who hold themselves to a high bar on correctness and maintainability, and a Solutions Architect who has already turned the workflow as it actually runs into a clear, buildable spec before it reaches you. The expectation is that you can take a well-scoped requirement and run with it, asking the right questions, not waiting to be directed.

The role grows toward owning the full AI architecture on a build: the model selection, the retrieval design, the evaluation strategy, and the decisions that get documented in the client's Blueprint. If you want to lead AI feature design and set the evaluation standard the team holds, this is the track that gets you there.

You are comfortable in TypeScript and Postgres; Python where the problem calls for it. You have shipped LLM-backed features that real people depend on, you think about failure modes and evaluation before the demo, and you can explain a model decision to a business owner without hiding behind technical language.