Codat Software Engineer (Data) at Codat building the data and reasoning layers behind an AI-first advisory intelligence platform. Responsible for building semantic, signal, and reasoning layers and running production systems.
Responsibilities
We're building the Advisory Intelligence Platform: Codat's client intelligence delivered inside the assistant a banker already uses (Microsoft Copilot first). Ask it anything about a client and it answers from the client's real books. Don't ask, and it still tells you what's changed, why it matters and what to do about it. Under it we're building the semantic layer, so every answer uses the same definitions, the signal layer that spots what's changed for a client, and the reasoning layer, a Codat-side LLM that turns those signals into advice. The first bank goes live this year. You'll be building the data and reasoning behind all of it.
This is a hands-on job for software engineers who want to build the systems a model reasons over, not configure a warehouse. Codat is AI-first, not AI bolted onto SaaS. The product lives inside the AI assistant our clients already use, and we build it the same way: agents do the repetitive work, engineers do the hard parts. You'll build it and you'll run it. Nobody hands it over the wall once it's live. You'll also get a real say in where it goes next. That means making your case to engineers, product and the commercial team, and bringing them along with you.
The semantic layer. Measures, dimensions and definitions the model is allowed to use, so every answer it gives is the same answer. Yours to grow, persona by persona.
The signal layer. Detecting what's changed for a client and surfacing it before the banker asks. You'll be designing and building this.
The reasoning layer. A Codat-side LLM that turns signals into advice. Our calculations are the ground truth it reasons over. You own the calculations the model relies on, and the names, definitions and structure that let it pick the right one.
The agents and MCP (Model Context Protocol) servers behind the assistant. What tools it has, how they're built, how they fail, and the evals that prove they work.
Everything from raw source data to the number the assistant quotes: the pipelines, services and data models, with the testing and observability to prove the number is right and the evals that catch the model quoting the wrong one.
You'll Probably Have
Built production systems from scratch, writing the code yourself rather than configuring managed services. You think about interfaces, failure modes and how the next person will change your code. If we ask about the hardest system you've built, you'll have a story ready.
Served data to an LLM and seen where it goes wrong: ambiguous definitions, poor retrieval, no evals. You know why "the model will work it out" isn't a data strategy.
Worked with data along the way, and liked it. You may never have been called a data engineer. You've built things that move, shape and serve data, and you know how they break, and how a model breaks on top of them.
Solid C#, Python or another backend language. Your code comes with tests, and you leave things tidier than you found them.
Qualification
Preferred
Built production systems from scratch writing code rather than configuring managed services