SERVICES / PRIVATE RAG
A document assistant has to do more than find a similar paragraph. It should retrieve the right source, respect who is allowed to see it and recognise when the evidence is not enough to answer.
01 / SOURCES
We look at what the team needs to ask and which documents can support a useful answer: formats, updates, versions, access and information that should stay out of the system.
The aim is to make an answer inspectable alongside its source, not merely plausible.
02 / TESTING
Putting a document into an index must not make it available to everyone. The approach depends on the source systems and their access model, including removal and investigation of errors.
Evaluation can use representative questions, expected answers, citations, abstention and checks across user roles. We agree the method around the data and scope.
03 / OPERATIONS
RAG can use cloud services, a hybrid design or components in a controlled environment. We compare them against sensitivity, integration, cost and operational responsibility.
“Private” needs a precise architecture; the label alone does not establish zero data movement or regulatory compliance.
ONE DOCUMENT TASK
Tell us which documents people rely on, who should be able to use them and what a good answer would look like. We can discuss what to test.
Discuss a RAG use case