Building AI Graphs With Rivet and MongoDB Atlas Vector Search to Power AI Applications

1
{
2
  "name": "default",
3
  "type": "vectorSearch",
4
  "fields":[
5
    {
6
      "type": "vector",
7
      "path": "embedding",
8
      "numDimensions": 1536,
9
      "similarity": "dotProduct"
10
    }]
11
}

1
Database: search
2
Collection: products
3
Path: embedding
4
K: 1

1
// Define a new POST endpoint for handling AI-enhanced search with Rivet
2
app.post('/aiRivetSearch', async (req, res) => {
3
4
    // Connect to MongoDB using a custom function that handles the connection logic
5
    db = await connectToDb();
6
7
    // Extract the search query sent in the POST request body
8
    const { query } = req.body;
9
10
    // Logging the query and environment variables for debugging purposes
11
    console.log(query);
12
    console.log(process.env.GRAPH_ID);
13
    console.log("Before running graph");
14
15
    // Load the Rivet project graph from the filesystem to use for the search
16
    const project = await loadProjectFromFile('./server/ai_shop.graph.rivet-project');
17
18
    // Execute the loaded graph with the provided inputs and plugin settings
19
    const response = await runGraph(project, { 
20
        graph: process.env.GRAPH_ID,
21
        openAiKey: process.env.OPEN_AI_KEY,
22
        inputs: {
23
            input: {
24
                type: "string",
25
                value: query
26
            }
27
        },
28
        pluginSettings: {
29
            rivetPluginMongodb: {
30
                mongoDBConnectionString: process.env.RIVET_MONGODB_CONNECTION_STRING,
31
            }
32
        }
33
    });
34
35
36
    // Parse the MongoDB aggregation pipeline from the graph response
37
    const pipeline = JSON.parse(response.result.value);
38
39
    // Connect to the 'products' collection in MongoDB and run the aggregation pipeline
40
    const collection = db.collection('products');
41
    const result = await collection.aggregate(pipeline).toArray();
42
    
43
    // Send the search results back to the client along with additional context
44
    res.json({ 
45
        "result": result, 
46
        "searchList": response.list.value, 
47
        prompt: query, 
48
        pipeline: pipeline 
49
    });
50
51
});