Time since we first saw it
Day 2
Similar jobs usually come down between day 2 and day 14.
From 956 similar postings that came down. Coming down is not always a hire.
United Kingdom
Actively hiring, first seen by us 2 days agoLocation type: RemoteFrom the employer’s posting
Employment type: Full-timeFrom the employer’s posting
Department: EngineeringFrom the employer’s posting
Design the reasoning layer: model context, tool use, and step-by-step question handling
Build eval sets and harness, and analyse whether changes actually improve answers
Choose which model handles which job and manage the cost per answer
Work on the semantic and signal layers with data engineers
Build and trace the agents and MCP servers the assistant calls
What goes into the model's context, which tools it calls, and when it should break a question into steps.
You'll build the eval sets and harness, and do the analysis that tells us whether a change made the assistant better or just different.
Which model should handle which job, and what that costs us per answer.
You'll work on both with the data engineers.
This covers the tools the assistant can call, including the ones that will eventually let it draft that email.
Probably Have Shipped LLM features that people actually used
Built evals for your own work
Good engineering habits in Python or another general-purpose language. That means tested, modular services rather than notebooks.
Strong SQL as well
some experience with CI/CD, Docker, cloud (we're on Azure) and tracing for LLM systems
Codat is an advisory intelligence solution purpose-built for modern commercial banking.
In the employer’s words · Read in context · Company website ↗
Actively hiringFirst seen by us 2 days ago, still listed, and not a repost of an earlier posting.
These dates show when we found the posting. Check the employer’s website to confirm it is still accepting applications.
Day 2
Similar jobs usually come down between day 2 and day 14.
From 956 similar postings that came down. Coming down is not always a hire.
4 open jobs
Open jobs by the week we first saw them, last 12 weeks.
Hiring recordAdd your resume and we’ll mark which of these tools you already use, and point out anything in the posting that could rule you out.
- Opinions about models that come from using them. You've chosen between models on how well they did and what they cost, and you know when a classical model would do a better job. - Good engineering habits in Python or another general-purpose language. That means tested, modular services rather than notebooks. Strong SQL as well, and some experience with CI/CD, Docker, cloud (we're on Azure) and tracing for LLM systems. - Led work without having the title. People followed an approach you set because you made the case with evidence, and you can explain the trade-offs to someone who isn't technical.
- AI as part of how you work, for research and prototyping as well as code. - Useful, not essential: fintech, lending or accounting; Spark or Databricks; semantic layers or metric views; fine-tuning. Don't tick every box? Apply anyway. We're more interested in what you've built and how you knew it worked than in a perfect match.
Complete your application on jobs.ashbyhq.com. The employer’s form will show what is required.
Already applied? Track this application
Yes. We’ve followed one other Senior AI Engineer job there since November 2026. It is still up.
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.
Some of the biggest commercial banks in the world already work with us. Their relationship managers would rather spend time with clients than dig through the books to find out what's changed. We're building this for them.
We're building the Advisory Intelligence Platform. It puts Codat's client intelligence inside the assistant a banker already uses, starting with Microsoft Copilot. A banker can ask it anything about a client and get an answer from that client's actual books. If they don't ask, it tells them anyway when something important changes, and suggests what to do.
It's a senior role and a hands-on one. Bankers make real decisions based on what the assistant says, so impressing people in a demo isn't enough. It has to be right. Codat is AI-first, not AI bolted onto SaaS, and we work that way ourselves, with agents handling the repetitive work so engineers can focus on the hard problems. You'll build what you design and look after it once it's live, and you'll have a real say in where we take it.
The reasoning layer. What goes into the model's context, which tools it calls, and when it should break a question into steps. Every number it gives has to trace back to a source, and it needs to know when to say it doesn't know. That last part is harder than it sounds.
Evals. You'll build the eval sets and harness, and do the analysis that tells us whether a change made the assistant better or just different. When a banker corrects an answer, that correction becomes a new test case.
Model choice and token economics. Which model should handle which job, and what that costs us per answer. Caching or a smaller model is often enough. For some questions a banker needs the strongest model available, and that's worth paying for.
The semantic and signal layers. You'll work on both with the data engineers. That means the definitions the model is allowed to use, and the signals that spot when something changes for a client. How the model uses them is down to you.
The agents and MCP (Model Context Protocol) servers behind the assistant. This covers the tools the assistant can call, including the ones that will eventually let it draft that email. You'll work out how they break, and add the tracing we need to see what happened inside an answer.
Classical modelling where it beats an LLM. Anomaly detection, peer benchmarks, classifiers. Sometimes the best answer is a simple statistical model, not an LLM.
Shipped LLM features that people actually used, and been the person they came to when something broke. You've designed how a model works with tools or structured data, and there are things you'd do differently now.
Built evals for your own work, and probably had one mislead you. We'd like to hear what you changed.
A feel for numbers, whether that comes from a quantitative degree or from your work. You want the denominator before you believe a percentage, and you notice when a headline figure is really about the mix of the data.
Opinions about models that come from using them. You've chosen between models on how well they did and what they cost, and you know when a classical model would do a better job.
Good engineering habits in Python or another general-purpose language. That means tested, modular services rather than notebooks. Strong SQL as well, and some experience with CI/CD, Docker, cloud (we're on Azure) and tracing for LLM systems.
Led work without having the title. People followed an approach you set because you made the case with evidence, and you can explain the trade-offs to someone who isn't technical.
AI as part of how you work, for research and prototyping as well as code.
Useful, not essential: fintech, lending or accounting; Spark or Databricks; semantic layers or metric views; fine-tuning.
Don't tick every box? Apply anyway. We're more interested in what you've built and how you knew it worked than in a perfect match.
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.
We file this job under AI Engineer, by the work the posting describes.
AI engineers build products on top of AI models that already exist, most often large language models: they connect a model to a company's data and tools and ship the assistants, agents and search features people use.
Postings ask for strong Python or TypeScript, experience with model APIs, retrieval and agent frameworks, and ways to test whether the model's answers are good enough.
An ML engineer trains and serves models; an AI engineer usually starts from a model someone else trained and builds the product around it, though titles often mix the two.
Selected passages from the saved posting. Check the full description for conditions and exceptions.
No pay amount identified in the saved description.
United Kingdom
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.
No clear work-authorization passage found. Eligibility is unconfirmed.