Knowledge assistants (RAG)
Ask a question and get an answer drawn from your manuals, policies, and past work, with links to the source so people can check.
Generative AI and LLM apps
Put large language models to work on your own documents and data.
Overview
Most companies already have the answers, buried in PDFs, emails, manuals, and the ERP. We build applications that find and use them: search that answers in plain language, assistants that draft from your templates, and summaries of records nobody has time to read.
What you get
- Chat and search over your own documents
- Drafting assistants for quotes, emails, and reports
- Model choice across OpenAI, Claude, Gemini, and open models
- Evaluation sets that measure accuracy
What we offer
Generative AI and LLM apps, in detail
Pick the parts you need. We scope them together, quote a fixed price, and build in two-week steps.
Drafting and summarising
First drafts of quotations, proposals, reports, and replies, written in your format for a person to review and send.
AI features in your product
Smart search, auto-fill, classification, and summaries added to software you already have.
Model selection
The right model for the job and budget, from OpenAI, Anthropic, Google, or open models you can host yourself.
Private deployment
Enterprise APIs that do not train on your data, or models running in your own cloud when data must not leave it.
Evaluation and monitoring
A test set of real questions with known answers, so accuracy is measured before launch and tracked after it.
Where it fits
Typical projects
- A sales assistant that answers product and pricing questions from your catalogue
- A policy assistant for HR and operations staff
- Summaries of long customer or patient histories
- Drafting tender responses from past submissions
Technology
What we use
Well-supported tools with a long runway, chosen for the project rather than for novelty.
- OpenAI
- Claude
- Gemini
- Llama
- LangChain
- LlamaIndex
- pgvector
- Python
How it works
From first call to production
Every engagement follows the same shape, so you always know what happens next and what you will see.
- 01 · Week 1
Scope
We sit with the people who will use the thing. You get a written scope, a fixed estimate, and an honest note about what we would cut first.
- 02 · Weeks 2 to 10
Build
Two-week iterations against a shared board. You see working software every fortnight, not a status deck.
- 03 · Launch week
Launch
Staged rollout, real monitoring, and a rehearsed way back. We stay on the call until the graphs go flat.
- 04 · Ongoing
Run
Patching, alerts, and a monthly review of cost and reliability. Leave with your code and your documentation whenever you want.
Questions
Frequently asked questions
Will our data be used to train public models?
No. We use enterprise API terms that exclude training on your data, or host open models in your own cloud when you need full control.
What if the AI gives a wrong answer?
We design for it. Answers show their sources, low-confidence answers say so, and important outputs go to a person for approval. We measure accuracy on your own questions before launch.
Which model do you use?
Whichever works best for your task and budget. We test a few on your data and often mix a large model for hard cases with a smaller, cheaper one for routine work.
Related
More AI services
- AI agents and workflow automationSoftware that reads, decides, and acts across your systems, with a person in charge.Learn more
- Document processing and OCRInvoices, purchase orders, and forms read automatically into your systems.Learn more
- AI assistants and chatbotsAnswer customers on your website and WhatsApp, day and night, from your own information.Learn more
Have a project like this in mind?
Send a paragraph about what you need. An engineer replies within one working day, and you get a written scope with a fixed price within a week.