Home / Signature cases / RAG on your archive
Signature · RAG on your archive
Ask your archive, get a sourced answer.
Technical manuals, FCA standards, environmental permits, internal policies: thousands of pages nobody has time to read. We build retrieval-augmented chat that gives staff and auditors a fast answer, with a citation for every claim and nothing invented.
← All signaturesSymptoms
The pain you are likely feeling.
- Compliance staff spend half their week answering the same questions about regulations, standards or permits, looking up the same PDFs every time.
- Auditors ask for the source of a decision and you need three days to find it across SharePoint, network drives and a legacy DMS.
- Generic GPT-style chat gives confident answers that turn out to be wrong, and you cannot defend that to a regulator.
- Your knowledge is locked in thousands of documents and only a handful of people know which document is the right one.
- A junior with a question wastes a senior’s afternoon, every week, on questions that already have documented answers.
Approach
How we approach it.
- Catalog the corpus.Which documents are authoritative? Which are deprecated? Which need OCR? Two weeks of document discovery before we touch the AI.
- Build retrieval that respects structure.Sections, headings, page numbers, references. We preserve them so every answer cites the source paragraph, not just the file name.
- Sourced answers, every time.No hallucinations passed through silently. Every response includes the document, page and exact excerpt that backs it up. If the corpus has no answer, the assistant says so.
- Microsoft Fabric or Databricks and Azure AI Foundry as the platform.Document storage in OneLake, embedding pipelines, governed vector search, Copilot Studio for the interface, all under one Microsoft contract.
- Iterate with the users who actually ask the questions.Compliance staff, technical writers, customer support. They test, we tune, until the answers are auditor-grade.
Proof
Where this signature landed.
SECO SecopediaRetrieval over technical certification standards, used daily by SECO’s auditors and clients.
OVOCOMQ&A over feed safety certification documents, scoring answers against the FCA framework.
SPW EnvironnementEnvironmental permit archive turned into a searchable, citable assistant for regional administrators.
Deliverables
What we typically ship in 90 days.
- Indexed corpus of your authoritative documents (typically 1000-10000 pages) with version control and a retire-old-versions workflow.
- Retrieval pipeline with hybrid search (keyword + vector), filters per audience, and consistent citations on every answer.
- Conversational interface in Microsoft Teams or a dedicated web portal, your users land where they already work.
- Quality scoring against a set of expected-answer test cases. We don’t go live until accuracy is auditor-defensible.
- Adoption pack for the team that owns the corpus: how to add documents, how to retire old versions, how to measure usage.
Sitting on an archive your team keeps looking up by hand?
Bring us your messiest documents folder. We’ll show you what auditable RAG on it actually looks like, and what it would take to put it live in 90 days.
Book a discovery call