Colocando o RAG em produção com o chatbot de IA da documentação do MongoDB

1
User: how to create a cluster atlas
2
Chatbot: To create a MongoDB Atlas cluster using the CLI, you can use the `mongocli atlas quickstart` command...
3
User: how to with UI
4
Chatbot: I'm sorry, I don't know what you mean...

1
User: how to connect to mongodb
2
Chatbot: To connect to MongoDB you can use the Java Driver...

1
User: write data to mongodb node.js
2
Chatbot: /* Reasonable generated answer */
3
    
4
Further reading:
5
- https://www.mongodb.com/pt-br/docs/drivers/node/current/usage-examples/insertOne/ (👍)
6
- https://www.mongodb.com/developer/languages/javascript/node-connect-mongodb/ (👍)
7
- https://www.mongodb.com/developer/products/realm/realm-meetup-javascript-react-native/ (🤷)

1
ingest pages --source docs-atlas

1
ingest embed --source docs-atlas \
2
  --since 2023-11-07 # only update documentation changed since this time

1
### Procedure
2
3
<Tabs>
4
5
<Tab name="MongoDB Atlas">
6
7
#### Go to the Search Tester.
8
9
- Click the cluster name to view the cluster details.
10
11
- Click the Search tab.
12
13
- Click the Query button to the right of the index to query.
14
15
#### View and edit the query syntax.
16
17
Click Edit $search Query to view a default query syntax sample in JSON (Javascript Object Notation) format.

1
---
2
tags:
3
  - atlas
4
  - docs
5
productName: MongoDB Atlas
6
version: null
7
pageTitle: How to Perform Semantic Search Against Data in Your Atlas Cluster
8
hasCodeBlock: false
9
---
10
11
### Procedure
12
13
<Tabs>
14
15
<Tab name="MongoDB Atlas">
16
17
#### Go to the Search Tester.
18
19
- Click the cluster name to view the cluster details.
20
21
- Click the Search tab.
22
23
- Click the Query button to the right of the index to query.
24
25
#### View and edit the query syntax.
26
27
Click Edit $search Query to view a default query syntax sample in JSON (Javascript Object Notation) format.

1
{
2
	_id: new ObjectId("65448eb04ef194092777bcf6")
3
	chunkIndex: 4,
4
	sourceName: "docs-atlas",
5
	url: "https://mongodb.com/pt-br/docs/atlas/atlas-vector-search/vector-search-tutorial/",
6
	text: '---\ntags:\n - atlas\n - docs\nproductName: MongoDB Atlas\nversion: null\npageTitle: How to Perform Semantic Search Against Data in Your Atlas Cluster\nhasCodeBlock: false\n---\n\n### Procedure\n\n<Tabs>\n\n<Tab name="MongoDB Atlas">\n\n#### Go to the Search Tester.\n\n- Click the cluster name to view the cluster details.\n\n- Click the Search tab.\n\n- Click the Query button to the right of the index to query.\n\n#### View and edit the query syntax.\n\nClick Edit $search Query to view a default query syntax sample in JSON (Javascript Object Notation) format.',
7
	tokenCount: 151,
8
	metadata: {
9
		tags: ["atlas", "docs"],
10
	    productName: "MongoDB Atlas",
11
	    version: null,
12
        pageTitle: "How to Perform Semantic Search Against Data in Your Atlas Cluster",
13
        hasCodeBlock: false,
14
	},
15
	embedding: [0.002525234, 0.038020607, 0.021626275 /* ... */],
16
	updated: new Date()
17
};

1
$filter

1
---
2
programmingLanguages:
3
  - shell
4
mongoDbProducts:
5
  - MongoDB Server
6
  - Aggregation Framework
7
---
8
What is the syntax for filtering data in MongoDB?

1
/**
2
  You are an AI-powered API that helps developers find answers to their MongoDB
3
  questions. You are a MongoDB expert. Process the user query in the context of
4
  the conversation into the following data type.
5
 */
6
export interface MongoDbUserQueryPreprocessorResponse {
7
  /**
8
    One or more programming languages present in the content ordered by
9
    relevancy. If no programming language is present and the user is asking for
10
    a code example, include "shell".
11
    @example ["shell", "javascript", "typescript", "python", "java", "csharp",
12
    "cpp", "ruby", "kotlin", "c", "dart", "php", "rust", "scala", "swift"
13
    ...other popular programming languages ]
14
  */
15
  programmingLanguages: string[];
16
17
  /**
18
    One or more MongoDB products present in the content. Which MongoDB products
19
    is the user interested in? Order by relevancy. Include "Driver" if the user
20
    is asking about a programming language with a MongoDB driver.
21
    @example ["MongoDB Atlas", "Atlas Charts", "Atlas Search", "Aggregation
22
    Framework", "MongoDB Server", "Compass", "MongoDB Connector for BI", "Realm
23
    SDK", "Driver", "Atlas App Services", ...other MongoDB products]
24
   */
25
  mongoDbProducts: string[];
26
27
  /**
28
    Using your knowledge of MongoDB and the conversational context, rephrase the
29
    latest user query to make it more meaningful. Rephrase the query into a
30
    question if it's not already one. The query generated here is passed to
31
    semantic search. If you do not know how to rephrase the query, leave this
32
    field undefined.
33
  */
34
  query?: string;
35
36
  /**
37
    Set to true if and only if the query is hostile, offensive, or disparages
38
    MongoDB or its products.
39
   */
40
  rejectQuery: boolean;
41
}

1
{
2
  "type": "vectorSearch,
3
  "fields": [{
4
    "path": "embedding",
5
    "dimensions": 1536,
6
    "similarity": "cosine",
7
    "type": "vector"
8
  }]
9
}

1
export async function getVectorSearchResults(
2
  collection: Collection,
3
  vectorEmbedding: number[],
4
  filterQuery: Filter<any>
5
) {
6
  return collection
7
    .aggregate<WithScore<EmbeddedContents>>([
8
      {
9
        $vectorSearch: {
10
          index: "default",
11
          vector: vectorEmbedding,
12
          path: "embedding",
13
          filter: filterQuery,
14
          limit: 3,
15
          numCandidates: 30
16
        },
17
      },
18
      {
19
        $addFields: {
20
          score: {
21
            $meta: "vectorSearchScore",
22
          },
23
        },
24
      },
25
      { $match: { score: { $gte: 0.8 } } },
26
    ])
27
    .toArray();
28
}