Knowledge assistant pilot
A knowledge assistant answers questions from your own documents and shows where each answer comes from. In a pilot we build it on one well-defined document collection and measure how well it performs, before you commit to anything more.
Four to six weeks
- Week 1: Scope. Together with your specialists we choose the document collection and 20–40 test questions.
- Weeks 2–3: Setup. Loading the documents, access control and a simple interface for the test group.
- Week 4: Evaluation. We run the test questions and measure whether the assistant finds the right source and answers correctly.
- Weeks 5–6: Tuning and decision. Improvements where the evaluation shows weaknesses, and a recommendation on the next step.
What you get
- A working assistant for a test group.
- An evaluation report: how often the right source is found, how good the answers are, and the known weaknesses.
- A recommendation to continue, adjust or stop, with the reasoning.
Framework
- Scope
- One document collection and one defined test group
- Decision
- Continue, adjust or stop
- Price
- Set out in the quote after scoping
Platform and data
The pilot can run on caveauAI from Blue Note Logic, on Azure OpenAI with Azure AI Search, on AWS Bedrock, or on open-source software on your own server. We recommend what suits your documents, requirements and existing systems. See Platforms.
Your documents stay under your control. We sign a data processing agreement before handling any personal data, and hosting can be in the EU or another region you choose.
Do you have documents that are hard to find your way around?
See how we document the work and its limitations.