Applied AI / Capability
Knowledge assistants.
Answer questions from your approved documents, with sources shown and access rules respected.
Describe the task
When to bring us in
Sound familiar?
- New staff ask the same questions because the answers are buried in policy documents.
- Our procedures exist, but finding the right version takes too long.
- We like the idea of an assistant, but can’t risk it making answers up.
The cost of leaving it
New staff keep asking the same questions, and the answers depend on who they ask. An out-of-date procedure gets followed because the current one was too hard to find.
Why us
What we bring to knowledge assistants.
A person stays in charge
Models work on bounded tasks inside your systems, and anything that matters goes to a named person to confirm.
Tested on your examples first
We measure a model on real cases from your work before it reaches staff, so you see where it fails while the stakes are low.
We say when AI is the wrong tool
Not every task should be automated. When conventional software or a person is the better answer, we tell you.
What the work includes
Answers your team can trace back to the source.
We decide which documents the assistant may use and who may see each one. Answers are built from passages retrieved from that set, and every answer shows where it came from. When the documents don’t cover a question, the assistant says so instead of guessing.
Source selection
Agree which documents are approved, who owns them and how updates reach the assistant.
Retrieval and answering
Search across the approved set, with answers built from the passages found.
Citations and refusals
Every answer links to its source, and unsupported questions get a clear “not covered”.
Access control
Users only get answers from documents they are already allowed to read.
Use cases
Projects we take on.
A policy helper for care staff
Staff ask how a procedure applies and get the relevant section of the current policy, linked to the full document.
Procedures help for finance teams
Questions about approval limits or month-end steps answered from internal procedures, restricted by team.
Search across teaching material
Instructors find guidance across course notes and handbooks without remembering which file it sits in.
Our approach
How we build retrieval you can check.
A knowledge assistant uses retrieval-augmented generation (RAG): it finds the relevant passages in your documents, then answers from them alone. Most of the quality comes from the retrieval, so that is where we spend the effort.
Finding the right passages
- Hybrid search: keyword matching for exact terms, plus semantic search for meaning.
- A reranking step that puts the most relevant passages first.
- Documents split along their own headings and tables, not at arbitrary lengths.
- Re-indexing when documents change, so retired versions stop being cited.
Answering only from sources
- Every answer links to the passages it used.
- When the documents do not cover a question, the assistant says so instead of guessing.
- Instructions hidden inside documents are treated as text, never as commands.
Access and data
- Permissions applied in the search index at query time, so restricted passages are never retrieved.
- Each client’s documents and embeddings kept apart.
- Embeddings follow the same retention and deletion rules as the documents they came from.
Measuring it
- Retrieval tested separately: do the top passages contain the facts needed?
- Answers tested for faithfulness: does every claim come from a retrieved passage?
- End-to-end checks against real questions with agreed answers, rerun on every change.
Guidance we build on
- OWASP Top 10 for LLM Applications 2025 (LLM01 prompt injection, LLM08 vector and embedding weaknesses)
- NIST AI 600-1: Generative AI Profile
We follow this guidance in how we design and test. It is not a certification. Our approach to responsible AI.
What you get
Outputs you can review.
- Approved source list and update process
- Question set for evaluation
- Assistant integrated into an existing tool
- Source citations on every answer
- Evaluation results and known limits
Technologies we use for this
Where it matters
- Health & social care Staff can look up care policies and procedures with the source shown; decisions about a person’s care stay with people.
- Medical education Educators can draft revision questions or case summaries from approved teaching material, then review every item before learners see it.
- Fintech & professional services Statements and application forms can be read into your systems field by field, and any value below a set confidence goes to a reviewer.
What’s changing
Shifts to plan for.
July 2026
EU high-risk rules move to 2027
The AI Omnibus entered into force on 27 July 2026, moving obligations for Annex III high-risk systems, such as those used in employment, credit scoring and education, to 2 December 2027.
August 2026
EU AI transparency duties apply
Article 50 obligations, such as telling people when they are dealing with an AI system, apply from 2 August 2026.
February 2026
New UK rules on automated decisions
The Data (Use and Access) Act 2025 widened when significant decisions can be automated, provided people are told, can contest the outcome and can ask for a human review.
Process
How the work runs.
Clean up the sources
Duplicates, drafts and outdated documents are removed or marked, because the assistant will repeat whatever it is given.
Write test questions
People who know the material write real questions with expected answers, including some the documents can’t answer.
Tune the retrieval
We adjust how documents are split and searched until the right passages come back for the test questions.
Start with a pilot group
A small group uses it first, and their unanswered or wrongly answered questions shape the next round of fixes.
Cost
What shapes the estimate.
Document condition
Scanned, inconsistent or poorly structured files need preparation before they can be searched well.
Access rules
Mirroring detailed per-document permissions takes more work than one approved set everyone can read.
Where it appears
A standalone page is simpler than embedding the assistant in an intranet or chat tool.
Ways to work together
A first, fixed stage
Discovery, a prototype or a technical review with agreed outputs, so the next decision is made with something concrete in hand.
Delivery in phases
The build split into releases, each with a scope you review and approve before it starts.
Ongoing engineering
An agreed scope for support and improvements after launch, with response expectations written down.
Common questions
Before you get in touch.
Will it work with documents that are out of date?
It will answer from whatever it is given, so document ownership matters. We set up a way to add, update and retire sources as part of the work.
Can it sit inside tools we already use?
Often, yes. It can be added to an existing web application, intranet or chat tool, depending on what those systems allow.
What affects the effort?
The volume and format of documents, how access is controlled today, where the assistant needs to appear and the quality bar agreed for answers.