Data Scientist

Help build the AI and data platform changing how financial advice is delivered.

Reporting to - Chief AI Officer
Employment - Full time
Location - Remote within the UK, with travel for team and customer meetings when required

About Afternoon
Afternoon is building the operating system for modern financial advice firms. We are replacing fragmented systems and manual processes with one AI- and data-first platform that helps firms work faster, deliver better advice and serve more clients.
At the heart of Afternoon is a genuinely difficult technical problem. We connect investment data from 27 platforms with client records, documents, meetings and communications, then use AI to turn that information into dependable, source-attributed outputs. Firms can prepare for meetings, capture and structure conversations, maintain client records, create complex advice reports and communicate with clients in one connected system.
Our growth over the past year has been strong. We are working with a rapidly growing number of advice firms, have a live B2B partnership with a UK bank and are increasingly being recognised across the market. Afternoon won the lang cat's Advicetech Catwalk 2026, voted for by more than 150 advisers, and was named Professional Services Start-up of the Year for Scotland at the UK StartUp Awards. We were also selected for J.P. Morgan's Fintech Forward 2026 programme.
We are well capitalised, ambitious and entering an important stage of growth. The team is still small enough for every new hire to shape the product, technical direction and company we become, while solving problems that matter to a large and established industry.

The opportunity
This is an opportunity to join Afternoon at the point where the product is proven, momentum is building and there is still enormous scope to shape our approach to AI. We are looking for a production-focused data scientist to help Afternoon understand and use information held across documents, conversations, client records and investment platforms. You will work on applied NLP and machine-learning problems where accuracy, evaluation, provenance and usability matter as much as model capability.  This is not a research-only role and it is not about producing isolated proofs of concept. You will take problems from exploration and experimentation through to production, working closely with software engineers, product colleagues and financial-advice specialists. You will see customers use what you build and help establish the data-science foundations for the next stage of our growth. The role suits someone who enjoys messy real-world data, asks good questions and wants to build systems that users can rely on.

What you will do

• Build and ship NLP and machine-learning systems for extraction, classification, retrieval, summarisation and structured generation, from first experiment to production.
• Work with LLMs, smaller task-specific models and deterministic methods, choosing the right approach for each problem rather than defaulting to one model or provider.
• Pull structured, source-attributed information out of complex financial documents and unstructured text, where every output can be traced back to where it came from.
• Design the evaluation that makes shipping safe: datasets, metrics and frameworks that surface failure modes in development and catch them in production.
• Own the pipelines behind the models, covering preparation, experimentation, training or fine-tuning, inference and monitoring, along with the reproducibility and experiment tracking that make results trustworthy.
• Work directly with engineers, product colleagues, advice specialists and customers: turning their problems into testable hypotheses, integrating models into product workflows through reliable APIs, and explaining plainly what works, what does not and why.

What we are looking for

• Strong Python, and production code other people have had to maintain.
• NLP or machine-learning systems you have taken to production yourself, in some combination of transformers, classifiers, LLMs, named-entity recognition, embeddings and retrieval, built on pre-trained models and open-source tooling rather than from scratch.
• Evaluation you have designed rather than inherited, and a view on the trade-offs between model quality, latency, cost and maintainability.
• Work that has moved beyond notebooks into services other people depend on.
• The ability to explain a technical decision, and its limitations, to someone who is not an engineer.
• Initiative: comfort making progress through ambiguity, and the confidence to pull people in when you need them.

Also useful
None of this is required. Any of it would give you a head start:

• Structured extraction from documents that are as often scans and images as they are text, using OCR, layout-aware models or vision-language models.
• Fine-tuning with SFT or LoRA, and the judgement to know when it beats a better prompt or a smaller model
• Synthetic data, weak supervision or self-supervised approaches, for the many problems here where labelled data does not exist yet.
• Function calling, tool use and agentic workflows, where a model has to act rather than answer.
• Putting models behind APIs and running the Python services around them, in our case with FastAPI and Pydantic.
• Experiment tracking and observability run with enough discipline that you can say why one model version beat another months later.
• Financial services, or any regulated domain where traceability and human oversight are not optional.

What we offer

• 25 days of annual leave, plus UK bank holidays.
• A remote-first working environment, with regular opportunities to work together in person.
• Significant ownership and the opportunity to shape the product, technical approach and ways of working.
• Direct access to experienced company and technical leaders, with scope to grow as Afternoon grows.
• A benefits package that will continue to develop with input from the team.

How to apply
Please send your CV and a short note explaining what interests you about Afternoon and the problems you would like to help us solve. We welcome candidates whose experience does not match every point but who can demonstrate strong judgement, learning ability and relevant impact.