AI that knows your company

An ordinary chatbot answers from what it picked up during training, so it knows nothing about your company. RAG first finds the relevant passages in your documents and answers from those, with a link to the source. In practice the difference is simple: asked about your own policy, it answers with your policy rather than a generic line from the internet.

The documents, the vector database and the application can run in your own Azure tenant, in a private cloud or on-premise. The language model itself is a commercial API (OpenAI, Anthropic, Google) called in a mode that does not train on your content, with an EU region and a data processing agreement. So the model does not run on your side, but it does not keep your data either. If you need the data never to leave your perimeter at all, I will say so at the start. That is a different category of project, not a switch.

An answer has to stand on a retrieved source, and the citation sits right next to it, so anyone can check it in two clicks. When there is no supporting material, the assistant says so instead of inventing something. Before go-live we run it against a set of real questions from your company and write the result down as a number, not an impression.

The build price comes from the volume of data and the number of integrations. Running it is two items: hosting, and model usage based on the number and length of queries. An estimate of the monthly running cost is always in the quote, because without it there is no way to decide on a project like this.

A pilot over a defined set of documents three to six weeks. A full rollout with integrations and role-based access two to four months, depending on scope.

I deal with it in the design rather than at the end: what gets into the knowledge base at all, what is anonymised and who is allowed to ask about what. I sign an NDA and hand over the data processing documentation you need for your own audits.