A grounded knowledge assistant for a service organisation
An illustrative AI programme: policies, contracts and technical documentation made answerable through retrieval, with existing permissions and source citations kept intact.
By IT NET Solution
- Sector
- Business services (illustrative)
- Scope
- Document ingestion, retrieval, permissions, assistant delivery
- Platforms
- Indurant, Evoyse, identity provider, document repositories
- Engagement
- Discovery, pilot and production operations
The challenge
Answers to routine client questions lived in four places: a document repository, a contracts folder, a support knowledge base and the experience of a handful of senior colleagues. New staff spent their first months asking around instead of reading.
An earlier experiment with a general assistant had been withdrawn after it answered a question using a document the asking colleague was not entitled to see — a permissions failure that no amount of prompt tuning would have fixed.
The approach
- 01
Bound the corpus
The pilot covered three document families with clear owners and an agreed review cadence, rather than the whole repository at once.
- 02
Carry permissions into retrieval
Access rules from the source systems are applied at retrieval time, so the assistant can only surface content the user could already open.
- 03
Require citations
Every answer returns the passages it was built from, with document, version and location, so a reader can verify rather than trust.
- 04
Keep a human in the loop
Client-facing use stays supervised: drafts are proposed, a consultant reviews and sends, and the review is logged.
The solution
Indurant handles ingestion, chunking and semantic retrieval across the three document families, including their Hungarian and English versions. Evoyse provides a single controlled route to the models, so provider choice, cost limits and logging are configured in one place instead of inside every application.
The assistant is delivered inside the tools the teams already use. When a document is updated, the index follows within the same working day, and superseded versions are excluded from answers rather than ranked lower.
What changed operationally
- Routine policy and contract questions are answered with a source reference attached.
- Access rules are enforced by the retrieval layer, not by the honesty of a prompt.
- Model access, spend and behaviour are visible in one place for review.
- New document families are added through an owner, a review cadence and a permission mapping.
Topics
- Enterprise AI
- Retrieval
- Knowledge management
- Governance
Related reading
Retrieval without the pilot trap
A retrieval assistant that answers well in a demo can fail quietly in production. What to measure, what to bound, and where human review still belongs.
Indurant adds multilingual retrieval
Our enterprise knowledge platform now retrieves across Hungarian and English documents in the same query, preserving citations and the permission model of the source systems.
Governing model access across the enterprise
Teams adopting several AI providers quickly accumulate keys, costs and inconsistent behaviour. A gateway model for routing, logging and controlled change.
Working on something similar?
Describe the systems, the constraints and what needs to change. We will outline a realistic path and the trade-offs behind it.