Senior Software Engineer (AI/Data)

Codat | Financial technology

2h 10m ago

Worldwide Remote

Full Time

About Codat

Codat is an advisory intelligence solution purpose-built for modern commercial banking. Through rich, specialized data, forward-looking insights, and integrated workflows, Codat empowers banking teams to deepen their relationships, grow their revenue, and simplify their day-to-day work.

Founded in 2017 and backed by JPMorgan, PayPal, Amex, Plaid, and Shopify, Codat has successfully powered over 350,000 connections to business customers’ financial systems — and is trusted by industry leaders to turn scattered information into actionable, strategic advantages in real time, every time.

The Role

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.

What You'll Work On

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.
Depth in at least one of SQL, Spark, Databricks/Delta Lake, Dagster/Airflow/Temporal or dbt. Bonus if you've shaped any of them into something a model can query, through a metrics layer, metric views or an MCP server.
Time with CI/CD, Docker, cloud infrastructure (we're on Azure) and observability. Even better if you set them up for a team.
An interest in the business behind the data. If someone asks for a number the data can't support, you'll say so, and you'll build the guard-rail so the model says so too.
Treated your own workflow as something to engineer. AI has already taken over the repetitive parts of it, and you're the person others ask how to do the same.

How We Work

We're a small team with plenty to do. Things move fast: we'd rather ship something useful and improve it than polish it in private. What you build is yours from start to finish.

There’s no return-to-office mandate, and you can work hybrid or fully remote. We trust you to be around when the team needs you and to deliver what you said you would. What we don’t want is anyone staying online late to prove a point.

Required Skills

pythoncsharpsqlsparkdatabricksdockerazuredbt

Required Languages

🇬🇧 English

The Role

What You'll Work On

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.

Depth in at least one of SQL, Spark, Databricks/Delta Lake, Dagster/Airflow/Temporal or dbt. Bonus if you've shaped any of them into something a model can query, through a metrics layer, metric views or an MCP server.

Time with CI/CD, Docker, cloud infrastructure (we're on Azure) and observability. Even better if you set them up for a team.

An interest in the business behind the data. If someone asks for a number the data can't support, you'll say so, and you'll build the guard-rail so the model says so too.

Treated your own workflow as something to engineer. AI has already taken over the repetitive parts of it, and you're the person others ask how to do the same.

How We Work

We're a small team with plenty to do. Things move fast: we'd rather ship something useful and improve it than polish it in private. What you build is yours from start to finish.