AI development
We build software that uses language models and other machine learning where it earns its place, and we say so plainly when it does not. Most of our AI work is ordinary software engineering around a model: clean data in, checked answers out, and a way to see when something goes wrong.
That includes retrieval-augmented generation (RAG), where a model answers from your own documents and cites what it used, assistants that sit inside an existing tool, and automation that takes over a repetitive, rule-heavy task.
What we do
The work in this area.
- Retrieval-augmented generation (RAG)
- Question answering over policies, manuals, tickets or knowledge bases. We handle ingestion, chunking, embeddings, search, and answers that point back to their sources.
- LLM applications
- Assistants, summarisation, classification and drafting features built into your product or internal tools, with prompts, guardrails and logging that can be inspected.
- Workflow automation
- Reading incoming email or documents, extracting fields, routing work, and updating the systems you already use, with a person approving the steps that matter.
- Evaluation and monitoring
- Test sets for the questions your users actually ask, so changes to a prompt or model can be compared before they ship.
Typical engagements
Examples of how people ask for this.
- A search-and-answer tool over an internal document library
- A drafting or summarising feature added to an existing web application
- Automating intake: forms, emails or PDFs into structured records
- A review of an existing prototype: what works, what is risky, what to change
How it works
From first call to handover.
Step 1
Define the task
We write down what a good answer looks like and collect real examples before choosing any model.
Step 2
Prototype against real data
A small working version, evaluated on your examples, so you can judge quality early.
Step 3
Build for production
Authentication, access control, logging, cost limits and error handling around the model.
Step 4
Hand over
Documentation, the evaluation set and a walkthrough so your team can keep improving it.
Tech
What we usually use.
Python, TypeScript, PostgreSQL with pgvector, FastAPI, Hosted and self-hosted models, Docker.
If your environment already runs on something else, we will tell you what we can support.
Contact
Talk to us about ai development.
A short message is enough. Describe the situation and what you would like to happen.
Other services