LLM & RAG Systems
How many days of paid leave do new starters get?
25 days a year plus public holidays, rising to 27 after three years.
- Employee handbook, page 14
- Leave policy update, March
1 document skipped because you don't have access to it
Ask your files. Get the answer and its source.
Your policies, manuals, tickets and contracts hold the answers your team keeps asking for. We build AI that finds the right passage, answers in plain language, shows where the answer came from and only shows people what they're allowed to see.
The approach is called retrieval-augmented generation, or RAG: the AI looks up your content before it answers instead of relying on memory.
What we build
Knowledge assistants
Ask a question and get an answer with links to the source, across your wikis, drives, help desk and databases.
Custom LLM applications
Tools built around a language model: pulling data out of documents, automating document-heavy work or powering features in your product.
Search that understands
Find things by meaning, not just keywords, across documents, tables, slides and scans.
Access that matches yours
Answers follow the same permissions as your source systems, so nobody sees what they shouldn't.
Measured accuracy
We build a test set of real questions and track how often the answers are right, before and after launch.
Where teams use it
- Support. Your team and your customers get answers from manuals, past tickets and policies in seconds.
- Sales. Product facts, pricing rules and competitor notes on hand during calls and proposals.
- Compliance and legal. Answers with exact citations, and a record of who asked what.
- Engineering. Search across docs, runbooks and tickets to unblock people faster.
- HR and operations. Policy and process questions answered once, correctly, for everyone.
How a build works
From scattered documents to an assistant people trust, measured at every step.
- Connect your sources. We link your document stores and systems, and map who's allowed to see what.
- Prepare the content. Documents, tables and scans are read, cleaned and indexed so the right passage can be found.
- Ground the answers. The assistant answers only from what it finds, cites its sources and follows your rules.
- Measure and tune. We track answer quality on real questions and keep tuning as your content changes.
When an answer has to come from your own documents, RAG is the right tool. When it doesn't, we'll recommend something simpler.
Prefer we run it for you?
Our managed knowledge assistant is the same system, delivered as a service. We connect your sources, set it up, keep it accurate and handle the upkeep, for a one-off setup fee and a monthly subscription.
Pricing depends on how many sources, documents and people are involved, so we quote it after a short call.
Setup done for you
Sources connected, permissions mapped and a working assistant delivered where your team already works.
Kept accurate
Regular quality checks on real questions, with fixes whenever answers drift.
Kept current
New and changed documents are picked up automatically.
Looked after
Monitoring, updates and improvements handled by us, with a named person to talk to.
How it starts
- A short call about your sources, your users and your security needs.
- A written proposal with a fixed setup fee and a monthly price.
- Setup and testing on your real questions, then launch.
Recent work
Questions, answered
Anything else? Ask us directly. We reply to every message personally.
How is this different from ChatGPT?
General chatbots answer from what they learned on the public internet. A RAG system answers from your own documents, shows the source and respects your permissions, so its answers are specific and checkable.
Can it keep confidential documents private?
Yes. It follows the permissions in your source systems, so people only get answers from documents they're allowed to open. We can also deploy it in your own cloud account.
What can it connect to?
Common document stores, wikis, help desks, CRMs, cloud storage and databases. If a system can export data or has an API, we can usually connect it.
How do you measure accuracy?
We build a set of real questions with known answers, score the assistant against it, and keep tracking it after launch.
Should we build it or use the managed service?
Build it if you want to own and run it yourselves. Choose managed if you want it to work without your team maintaining it. We'll recommend one on the first call.
How much does it cost?
Builds are scoped and priced up front, starting with a fixed-price pilot. The managed service has a one-off setup fee and a monthly price, quoted after a short call.
Tell us what you want AI to do.
Someone senior reads every message and replies personally. Short notes are fine; we'll ask the right questions on the call.