Enterprise Knowledge Assistants
Large organisations do not have a knowledge problem so much as a retrieval problem. The policy exists; it is in a PDF on a SharePoint site nobody remembers, superseded by a memo in an email thread.
We build assistants that search across those sources under the same access rules that govern the originals, answer with citations, and refuse rather than improvise when the corpus does not contain the answer. That last behaviour is usually what gets the system approved.
Platforms we typically deliver this on
Scope of the service
Source connection and ingestion
Pipelines from SharePoint, network shares, document management systems, wikis, ticketing systems and databases, with change detection.
Permission-aware retrieval
Query-time filtering against the source system's own entitlements, so the assistant can never surface material a user could not otherwise open.
Citation and traceability
Every answer links to the passages it used, with document version and date so staff can verify before acting.
Currency and supersession handling
Detection of outdated and superseded documents so the assistant answers from the current policy rather than whichever version indexed best.
Refusal and escalation behaviour
Explicit handling for questions the corpus cannot answer, with routing to the team that owns the topic.
Usage analytics
What people ask, what the assistant could not answer, and where your documentation has genuine gaps.
Where this is used
Representative engagements, described at the level our clients permit. Sector and shape are accurate; identifying detail is withheld.
Policy answers for front-line student services
Advisors searched across academic regulations, financial aid rules and procedure manuals held in four systems. The assistant answers from the current approved version and cites the clause.
Outcome — Consistent answers across advisors, with the citation available if a student challenges one.
Clinical policy and procedure lookup
Staff needed current procedure without leaving the ward. Retrieval is filtered by role and site so the answer reflects local protocol, not another region's.
Outcome — Current, site-correct procedure retrieved in seconds instead of minutes of searching.
Preparing responses to information requests
Records teams use the assistant to locate responsive material across archives, with a complete log of what was searched and retrieved as part of the response file.
Outcome — Search effort reduced and the audit trail generated as a by-product.
Before you get in touch
Can staff see documents they should not?
No. Retrieval filters against the source system's permissions at query time rather than maintaining a separate copy of the access model, so there is nothing to drift out of sync.
What happens when a policy is updated?
The pipeline detects the change and re-indexes. Superseded versions are marked so answers come from the current document, and the assistant can tell a user when something recently changed.
How do you stop it inventing answers?
It answers only from retrieved passages, cites them, and is evaluated specifically on its refusal behaviour for out-of-corpus questions.
Often engaged alongside this
Custom AI & Generative AI Development
We build the AI systems that do not exist off the shelf: document processing against your forms, classifiers trained on your taxonomy, copilots that read your systems of record.
Read moreAIAI Architecture & Platform Engineering
Model choice is the easy part. We design and build the platform layer — data access, identity, evaluation, observability, cost control — that everything else depends on.
Read moreData & AnalyticsData Governance & Quality
Catalogues that are current, quality rules that run automatically, and ownership that maps to how your organisation is genuinely structured.
Read moreStart with three weeks and a straight answer
The AI Readiness Assessment is fixed in scope, fixed in price and produces four deliverables you own — whether or not you continue with us.