README.md

metadata

title: Mongodb Gemini Rag
emoji: ♊️
colorFrom: indigo
colorTo: purple
sdk: gradio
sdk_version: 4.31.5
app_file: app.py
pinned: false
license: apache-2.0

Atlas Vector Search Chat with MongoDB and Google Gemini

Welcome to the Atlas Vector Search Chat! This application demonstrates how to use MongoDB Atlas Vector Search with Google Gemini for semantic search and retrieval tasks.

Features

• Interactive Chat: Ask questions related to the embedded documents.
• Vector Search: Utilizes MongoDB Atlas Vector Search to find relevant documents based on similarity.
• Google Gemini Integration: Embeds text and generates responses.

Requirements

• Python 3.7 or later
• MongoDB Atlas account
• Atlas cluster enabled with 0.0.0.0/0 connection and connetion string
• Google Cloud account with access to Gemini

Installation

1. Clone the space:

   • Click [...] and clone the space to your repo, make sure to input the variables:

2. Set up environment variables:

   • GOOGLE_API_KEY: Your Google API key for Gemini.
   • MONGODB_ATLAS_URI: Your MongoDB Atlas connection string.

Running the Application

1. Start the application:

   python app.py
   

2. Access the interface: Open your browser on the App tab.

Vector Search Index Configuration

To create a vector search index on the google-ai.embedded_docs collection, use the following configuration:

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

MongoDB Trigger to Embed Results

This Atlas enviroment use an Atlas Database trigger on collection google-ai.embedded_docs to capture any insert operation and embed the content as specified in this article.

// Get the API key from Realm's Values & Secrets
const apiKey = context.values.get('google-api-key');

// Set up the URL for the Google Generative Language API - embedding endpoint
const url = `https://generativelanguage.googleapis.com/v1beta/models/embedding-001:embedContent?key=${apiKey}`;
// batch example
// const url = `https://generativelanguage.googleapis.com/v1beta/models/embedding-001:batchEmbedContents?key=${apiKey}`;

// Get the full document from the change event
const doc = changeEvent.fullDocument;

try {
    console.log(`Processing document with id: ${doc._id}`);

    // Prepare the request body
    const requestBody = `{
    "model": "models/embedding-001",
    "content": {
    "parts":[{
      "text": '${doc.content}'}]}}`;
    

    // Make the HTTP POST request
    const response = await context.http.post({
        url: url,
        headers: { 'Content-Type': ['application/json'] },
        body: requestBody
    });

    // Parse the JSON response
    const responseData = EJSON.parse(response.body.text());
    console.log(JSON.stringify(responseData))

    if(response.statusCode === 200) {
        console.log("Successfully received embedding response from the API.");

        // Extract the embedding from the response
        const embedding = responseData.embedding.values; // Adjust based on actual response structure

        // Use the name of your MongoDB Atlas Cluster
        const collection = context.services.get("mongodb-atlas").db("google-ai").collection("embedded_docs");

        // Update the document in MongoDB with the embedding
        const updateResult = await collection.updateOne(
            { _id: doc._id },
            { $set: { embedding: embedding }}
        );

        if(updateResult.modifiedCount === 1) {
            console.log("Successfully updated the document.");
        } else {
            console.log("Failed to update the document.");
        }
    } else {
        console.log(`Failed to receive embedding. Status code: ${response.statusCode} -  ${JSON.stringify(response)}`);
    }
} catch(err) {
    console.error(`Error making request to API: ${err}`);
}