Simple_RAG_app

To start add you openai_api_keyin the secrets.toml file located in .\streamlit folder Then, just run streamlit run streamlit_app.py

Purpose

The aim is to provide a simple app implementing a RAG function on specific document. It provides a simple Chroma database abstraction layer combined with an simple chat agent to perform RAG

Structure

├── src
│   ├── .streamlit
│   ├── Database
│   │   ├── Private
│   │   │   ├── Chroma_DB
│   │   │   ├── Files
│   │   ├── Public
│   │   │   ├── Chroma_DB
│   │   │   ├── Files

Adding a new document

New document can be added using the following methods

• Adding the document to \Files folder
• Calling the function add_missing_document_to_chroma_database or using the button Add missing files to DB in the Chroma DB management expander

Prompt

The prompt can be changed in the app LLM paremeters expander.

Working with multiple database

Defaut option is using only data from public database Data from private database can be also used, in this case the app loop through both DB

working with retrieval agent or tool agent

The POC offers the possibility to use both a retrieval agent or a tool agent. It is possible to switch using the Use agent checkbox in App LLM parameteres expander Differences are the following:

• retrieval agent can only perform RAG on given document. It returns good quality answer.
• tool agent is able to determine and use different tools and provide a custom answer. It can currently use the retrieval agent / a tool to do math / its own knowledge database. Answer quality seems to be a little bit lower on retrieval (way of improvment)