rag.md

Meeting-minutes RAG

With our final content collection set up, we can make use of keyword search and vector search.

Configuring search

Navigate to your Atlas database and to the "Atlas Search" tab of your cluster. We need to create two indexes:

• A text search index called "text_index"
• A vector search index called "vector_index"

Search configuration:

{
  "mappings": {
    "dynamic": true,
    "fields":{
      "organizations":[
        {
          "type": "autocomplete",
          "tokenization": "edgeGram",
          "minGrams": 1,
          "maxGrams": 10
        },
        {
          "type": "token"
        }
      ],
      "user_id":[
        {
          "type": "token"
        }
      ]
    }
  }
}

Vector search configuration:

{
  "fields": [
    {
      "type": "vector",
      "path": "embedding",
      "numDimensions": 768,
      "similarity": "cosine"
    },
    {
      "type": "filter",
      "path": "organizations"
    },
    {
      "type": "filter",
      "path": "user_id"
    }
  ]
}

Running the remo app

Start the search demo app by running

python -m streamlit run meeting_mate/rag/streamlit_rag.py

This should launch our demo app in the browser