Custom AI development that runs inside your business
Some businesses need a new AI product built from scratch. Many more need the software they already run to do more: answer questions from their own documents, read the invoices that arrive by email, draft the reply to a customer, or find the record someone is looking for. We do both. We build standalone AI applications and agents, and we build AI into your apps, websites, portals, stores, CRM and other business systems, on a model you control, measured against a result you agreed before we started.
- Standalone AI products
- Assistants over your documents
- Document AI
- AI features inside your existing apps
- Retrieval systems (RAG)
What we build with AI
Standalone AI products
AI applications built from scratch: a customer-facing assistant, an internal knowledge tool, or a new product with AI at its core, with the backend, interface and hosting included.
Assistants over your documents
Ask a question in plain English and get an answer with the source cited, from contracts, manuals, past projects or policies.
Document AI and vision
Invoices, purchase orders, forms, scans and photos read automatically, with the data placed in your accounting, ERP or CRM and a review queue for anything uncertain.
AI features inside your existing apps
Semantic search in a portal, summaries on every support ticket, recommendations in a store, or an assistant inside a mobile app.
Retrieval systems (RAG)
The plumbing that lets a model look things up in your data before it answers: ingestion, chunking, search, citations and permissions.
Fine-tuning
When a general model needs to learn the language of your field or follow a strict format, we tune an open-weight model on your examples.
Who this is for
Companies that already have systems and data, and want AI to work inside them rather than beside them. Teams whose data cannot leave their own servers. Founders building an AI product from scratch, or adding an AI feature to a product that already has users. And businesses that have tried a chatbot, watched nobody use it, and want something that actually does a job.
How an AI project runs
Pick one workflow
We start with the task that costs your team the most time, not with the technology.
Outcome: a written description of the workflow and the result we are aiming for.
Check the data
We look at where the information lives, how clean it is, and whether it can leave your environment.
Outcome: a decision on private or cloud models and a list of what needs connecting.
Build a working version
The model, the connection to your system and a simple interface, running on real data.
Outcome: something your team can try.
Test with the people who will use it
A small group runs it on real work while we tune it.
Outcome: a measured result against the baseline.
Roll out and hand over
Documentation, monitoring, a review queue for the model’s uncertain cases, and training.
Outcome: a system your team owns and can see into.
How long each step takes depends on your data and how many systems are involved. We set expectations at the first call and update them every week.
Models and tools we work with
Open-weight models such as Llama, Qwen and Mistral, served with vLLM or Ollama when the data must stay private. OpenAI, Anthropic and Azure OpenAI through your own account when a frontier model is the right choice. LangGraph and n8n for agents and workflows. pgvector and Qdrant for search. Python, Node.js, Next.js and Laravel for the code around them. Your databases, apps, CRM, store, Microsoft 365 or Google Workspace as the data sources.
Where this leads
Questions people ask about AI projects
Something else on your mind? Ask an engineer directly.
Do we have to send our data to OpenAI?
No. For most business tasks an open-weight model running on your own server or in your cloud account is good enough. We use the big vendors’ models only when you want them, under your account, with retention switched off.
How do you know if it is working?
We agree a measure before we build, such as minutes saved per document or the share of questions answered correctly, and we test against it with real examples from your business.
What if it does not work well enough?
You get the written findings and the code either way. Sometimes the honest answer is that the data is not ready yet, and we tell you what would need to change.
Can you add AI to an app another company built?
Yes. We work in your repository, follow its conventions and write tests, in Python, Node.js, React, Next.js, Laravel, PHP, .NET, Flutter, React Native, Swift or Kotlin.
Who maintains it afterwards?
Your team with our documentation, or us under a support plan. Models improve quickly, so we recommend a quarterly review.
Tell us the task your team does by hand every day
Describe it in a sentence. An engineer will tell you whether AI can take it, whether it should run on your own servers, and what a first version would involve.