Services

Focused projects with a clear outcome. Most start with a short call and a fixed-price prototype, so you see something working before committing to a full build.

AI assistants and retrieval

Most organisations already have the answers written down: policies, manuals, past tickets, a website. We build assistants that search that material and answer in plain English, quoting the passage they used. Because answers come from your content rather than the model's memory, they stay accurate, and you can see exactly why an answer was given.

Typical uses: a staff helpdesk over internal policies, customer support over product documentation, or a public information service over official guidance.

What you get

  • Ingestion pipeline for PDFs, web pages and databases
  • Vector search with a citation on every answer
  • Chat interface on your site or as a standalone app
  • A test set, so accuracy is measured before launch
  • Admin controls for updating content

Built with

  • Python
  • FastAPI
  • Supabase
  • pgvector
  • Open-source embeddings
  • OpenAI or Anthropic models

AI-powered web apps and PWAs

A progressive web app runs in the browser but installs like a native app, works offline and doesn't need an app store. We build them for products that need AI features from the start: sign-in, subscriptions, a database and the AI service behind it, all in one codebase you own.

This is the same stack we use for our own products, so the boring parts (auth, payments, hosting, monitoring) are already solved and the budget goes on what makes your app different.

What you get

  • React front end that installs on phones and desktops
  • Supabase for sign-in, database, storage and Edge Functions
  • Stripe subscriptions, including free tiers
  • Hosting on Netlify with a preview of every change
  • Analytics and error monitoring from day one

Built with

  • React
  • Supabase
  • Edge Functions
  • Netlify
  • Stripe

Automation and content pipelines

Repetitive multi-step work is where automation pays for itself: gathering research, drafting content, generating reports, posting updates. We build pipelines that do the routine parts and stop for a human check before anything goes out.

Every pipeline keeps logs you can read and a review step you control. The aim is to take the grind out of the work, not the judgement.

What you get

  • Workflows connected to the tools you already use
  • A review step before anything is published
  • Scheduled runs with readable logs
  • Results fed back into the next run

Built with

  • Python
  • n8n
  • REST APIs
  • Instagram Graph API
  • Text-to-speech and video tools

Data, dashboards and models

Data sitting in spreadsheets and databases usually has more to say than anyone has time to find. We clean it, connect it and put it in front of people as dashboards and reports. Where it helps, we train models on it: classifying reviews, predicting demand, flagging outliers.

Every model comes with a plain-English note on how well it works and where it fails, so nobody has to take a number on trust.

What you get

  • Clean, documented data pipelines
  • Dashboards with live figures
  • Classification and prediction models with honest evaluation
  • A written summary of what the data does and doesn't support

Built with

  • Python
  • pandas
  • scikit-learn
  • SQL
  • Postgres

Public services and health

A lot of our own products sit around public services, from NHS waiting lists to benefits guidance. That is a deliberate choice of problem, not a past career. We build for people who are stressed, on a phone, and need a clear answer, and we design the retrieval so those answers can be checked against the source.

If you're a council, charity, advice service or health organisation, we can talk in your language from the first call, and we'll document data flows in the form your governance team needs.

What you get

  • Assistants grounded in official guidance, with citations
  • Data hosted in UK or EU regions
  • Documented data flows for your records
  • Accessible interfaces tested on real phones

How we work together

Three ways in, depending on how sure you are about what you need.

Free, 30 minutes

Discovery call

We talk through the problem and tell you honestly whether AI helps. If it doesn't, we'll say so.

Fixed price, one to two weeks

Prototype

A working version with your real data. Enough to put in front of users and decide whether to go further.

Quoted per project

Build and support

The production version, then optional monthly support with a fixed fee and a named contact.

Questions we get asked

Which AI models do you use?

Whichever fits the job and your budget. We work with OpenAI and Anthropic models through their APIs, and with open-source models where data needs to stay in-house. Retrieval and embeddings usually run on open models, which keeps running costs low.

Where does our data go?

Your documents live in a database you own, hosted in the UK or EU. Questions are sent to the model provider you choose, and we document exactly what leaves your systems so you can put it in front of your governance team.

Can you work with a small budget?

Yes. The prototype stage exists for this reason. It's fixed price and gives you something real to judge before spending more.

Can you work alongside our existing team?

Yes. We can build a standalone piece or work in your repository with your review process and your deployment setup.

How long does a typical project take?

Prototypes take one to two weeks. Full builds usually run four to ten weeks depending on scope, with something usable well before the end.

Do you support it after launch?

Yes, on a monthly basis if you want it. Everything is documented so you can also run it yourself or hand it to another developer.

Not sure which of these you need?

Describe the problem and we'll tell you the smallest thing worth building first.

Start a project