Local AIDocument search
An AI assistant to find your internal documents
When a quote, a procedure or a guide is scattered across several folders, a simple search can interrupt several people’s work. This project presents an assistant built into the CRM that helps teams find this information from the tool they work in. The model runs on the company’s own servers.
- Chatbot
- Internal documents
- In-house hosting

(01)Where work gets stuck
Before calling a client back, you have to find the right quote.
An employee needs to find a client’s latest quote before calling them back. They search the folders, check the attachments and sometimes ask a colleague which version to use.
(02)From question to document
Illustrative scenarioFour steps, from the question to the source.
Here is how the journey can run. It is an illustrative scenario: these features are not presented as a confirmed deployment.
- 01
Question
The employee asks their question from the CRM.
- 02
Search
The assistant searches the internal documents the employee has access to.
- 03
Source
It points to the relevant document and provides a link to open it and check the source.
- 04
Uncertainty
If several versions exist and the most recent one cannot be identified with certainty, it flags the uncertainty and asks the employee to check.
(03)Concrete example
Illustrative scenario“Where can I find the latest quote?”

From a client’s record, the employee asks this question. The assistant suggests the file “Devis.pdf” with its date and a link to the document. The employee can check the version before using it.
(04)Value for the business
Expected benefitsWhat teams stand to gain.
Less switching between tools
Teams can find a document without moving from one tool to another.
Fewer interrupted colleagues
They can ask colleagues less often where a file is or which version to use.
A source to check on the spot
They can check the source directly, from the link the assistant provides.
In-house hosting
In-house hosting lets the company keep control of the environment the model runs in.
(05)Indicators to track
To be measuredThree things to measure after deployment.
This page announces no measured results. These indicators will help judge the assistant once it is in service.
Average time to find a document
From the question to opening the right document.
Share of answers with a relevant source
Answers that point to the right document, out of all answers.
Searches that still need a colleague
The number of searches the assistant alone cannot resolve.
Services behind this project