Solution
Document search
Keyword search fails the moment the person searching doesn’t use the words the author wrote. Search by meaning finds the right document even when the vocabulary has moved on a decade.
The gains from search
The archive stops being dead weight
Years of reports, minutes and contracts become consultable again. That’s often the most immediate gain, before any generative AI comes into it.
Less work redone
The most expensive question in any organisation is “have we dealt with this before?”. Answering it in thirty seconds avoids doing it twice.
A sound base for what comes next
An assistant is worth exactly what the retrieval feeding it’s worth. Getting indexing right first avoids building on sand.
In practice
- An index, a search catalogue, built on the exact words and on the meaning of each passage, combining both sets of results
- Chunking suited to the document: a contract does not split like a set of minutes
- Text extraction from scanned documents, including older holdings
- Filters by date, type, department and confidentiality, applied before the search rather than after
- Results shown with their extract and their origin, so the full document is one step away
- Automatic reindexing when a document changes, without rebuilding the whole set
Systems involved
- File shares, SharePoint, Nextcloud
- Document management and municipal software
- PostgreSQL with vector search
- Scanned archives and PDF holdings
- Existing websites and intranets
Service lineArtificial intelligence →
The archives searched
The same work, against each sector’s own constraints. Every card opens the full sector.
Banking and insurance
Search across contracts and endorsements
Finding an obligation by its meaning rather than its exact wording, across documents that share neither structure nor vocabulary.
Energy and utilities
Search across plant documentation
Diagrams, test reports and operating procedures found by their content, including in scanned documents.
Healthcare and life sciences
Scientific monitoring
Tracking the literature of a field with a summary and a link to the source, as the laboratory demonstration already does.
Manufacturing
Search across shop-floor documentation
Routings, drawings and incident reports found from the problem as described, without knowing the document reference.
How it runs
Sample
A few hundred representative documents, to measure what the current search genuinely misses.
Indexing
Chunking, extraction and index built, then judged on real searches rather than on a laboratory score.
Interface
The search placed where people already work, rather than one more tool to open.
Follow-up
Searches that return nothing are the best signal of what the archive lacks. We read them, and fill the gaps.
To go deeper
A knowledge base the AI can read and a human can review
An organisation's knowledge lives in people's heads, in mailboxes and in a portal nobody opens any more. A folder of versioned Markdown files changes three things: review becomes possible again, authority becomes traceable, and the assistant stops guessing where one idea ends and the next begins.
RAG, loops, graphs: four decisions, not four products
The vocabulary of AI agents renews itself every quarter. Behind RAG, GraphRAG, context engineering, loops and graphs sit four distinct decisions: what the model sees, what verifies, who runs next, and where your signature goes.
Test the fit: Document search
Describe the context, constraints and decision you need to make. The first conversation qualifies scope, boundaries and the next useful step.
Describe the situation