Policy and procedure search
Help employees find relevant guidance and distinguish approved policy from supporting notes, while respecting access restrictions on sensitive material.
AI agents & knowledge
RAG & knowledge management
Help people find grounded answers in your approved documents and organisational knowledge.
Built around a specific business need.
Useful organisational knowledge is often spread across policies, manuals, shared folders and individual expertise. People may find several plausible answers without knowing which is current. Retrieval-augmented generation, usually shortened to RAG, combines document retrieval with generated responses so an assistant can use selected source material when answering a question.
Kent Consultancy would treat this as a knowledge management project as well as an AI implementation. We would identify authoritative sources, preserve their ownership and design retrieval around actual questions. Grounded answers still need evaluation: a retrieved passage can be outdated, incomplete or misinterpreted, and a source reference alone does not prove that an answer is correct.
Who this is for
- Organisations with substantial policy, technical or operational documentation.
- Internal teams spending time locating and reconciling information across repositories.
- Service owners who need answers linked to controlled, reviewable sources.
Where this can make a practical difference.
Technical knowledge assistance
Retrieve applicable product or maintenance information, preserve version context and show source passages so specialists can check the reasoning before acting.
Knowledge quality discovery
Use unanswered questions and conflicting retrieval results to identify missing guidance, duplicate documents and areas where ownership or review dates need attention.
How we approach the work.
Audit sources and permissions
Inventory repositories, document types and access rules. Identify authoritative versions and decide which material should be excluded before creating an index.
Prepare retrievable content
Extract text, structure content into useful sections and retain metadata such as title, version and owner. Check whether tables, scanned pages or diagrams require additional handling.
Design retrieval
Select an appropriate search approach and test how well it returns relevant evidence. Preserve user-level access controls through retrieval and handle documents that have changed or been withdrawn.
Compose source-aware answers
Instruct the answer stage to distinguish supported statements from missing evidence. Provide useful references and define when insufficient or conflicting information should trigger clarification or handover.
Evaluate and maintain
Test retrieval separately from answer quality using representative questions. Agree refresh schedules, deletion handling and an ownership process so the knowledge base remains useful after launch.
The sequence is tailored to the agreed scope, dependencies and risk. It describes our delivery approach, not a fixed timeline or performance guarantee.
What the engagement can deliver.
- A source inventory and knowledge ownership map.
- Content preparation and indexing workflow.
- Permission-aware retrieval and answer configuration.
- A reference question set with evaluation findings.
- Guidance for updates, exclusions and ongoing review.
The proposal identifies the deliverables, acceptance evidence, responsibilities and support arrangements included in your engagement.
Measure what matters to this workflow.
Retrieval relevance
Whether the required evidence appears in the returned source material.
Answer support
Reviewed answer statements justified by the cited source passages.
Knowledge freshness
Indexed material aligned with current approved source versions.
These are measures to discuss and baseline. Targets are agreed for the specific system and data; no universal improvement percentage is assumed.
What to prepare for discovery.
- Accessible source documents with identifiable owners and versions.
- Examples of questions that staff currently struggle to answer.
- Clear permissions, retention expectations and a process for retiring content.
Questions worth asking.
Does RAG mean training a model on our documents?
Usually the documents are prepared for retrieval and supplied as context when needed. That is different from changing a model’s learned parameters through training.
Will references remove incorrect answers?
No. References make checking easier, but the system can retrieve the wrong material or interpret it badly. Both retrieval and answer support need explicit evaluation.
What if documents disagree?
The system should follow an agreed authority and version policy, or flag the conflict. It should not silently merge contradictory guidance into a confident answer.