# AI that answers only from your own documents, and cites every source

Source: https://ai.prospicience.in/capabilities/rag-with-citations

We build private knowledge libraries that answer staff questions only from your approved documents, and name the document and page behind every answer. When the library holds no good match, the system says so instead of guessing. Our specialist education library went from brief to live in two weeks, with access controlled per user.

## What you get

- Answers come only from your approved documents, each with its source, so staff can act on them with confidence.
- The library is searched before the AI is ever asked. When nothing matches well enough, the system says it has no answer, so staff never receive an invented one.
- Citations are taken from the source passage itself, so every reference points at a page that really says what the answer claims.
- Documents are read as information, never as instructions, so a stray line inside a file cannot redirect the assistant.
- Access is set per person and enforced at the data layer, so each member of staff sees exactly what they are cleared to see.

## Systems we have shipped for this

- E02 A library that answers questions and always shows its source (Specialist education institute): https://ai.prospicience.in/work/knowledge-library-with-citations
- E14 An admissions assistant that answers in Gujarati, Hindi or English, by text or voice (Pre-school group): https://ai.prospicience.in/work/multilingual-admissions-assistant

## What businesses come to us for

- A hospitality and retail operator wanted four assistants inside their team chat, each trained on its own policies and standard operating procedures.
- A work management startup wanted a context engine that finds the right material for a model and prioritises it within a fixed budget.
- A pharmacy group wanted prescriptions read and entered into their system, with patient privacy protected throughout.

## Questions

### What is retrieval augmented generation in plain terms?

The AI answers from your documents rather than from memory. It searches your library first, pulls the passages that match, and answers using only those. That is what lets every answer show exactly where it came from.

### How do you stop an AI chatbot hallucinating about company policy?

We make "no answer" the default. The library is searched first, and if no passage is a strong enough match the system replies that it has no answer, without calling the AI at all. Invented answers come from asking a model a question with nothing to support it, and our design never does that.

### Can we keep the documents private?

Yes. Documents sit in private storage reached through short lived links, and who can see what is enforced per user at the data layer. Two people asking the same question each get the answer their access allows.

### How long does a knowledge library take to build?

Our specialist education library went from brief to live in two weeks, including a switch of search provider mid project. The biggest factor in your timeline is how settled your source documents are.

### What happens when our documents change?

Updated documents are indexed again and the answers follow straight away. There is no retraining step, so the moment a policy changes, the assistant quotes the new version.