Docs
Use case
Explained Prompt: Show me forecast for ZEKO
Authorization for RAG
Only retrieve documents users have access to. Avoid leaking data to a user that should not have access to it.
Scenario
When a user submits a forecast inquiry for a specific company, like Zeko, the chatbot will generate a response using relevant documents retrieved from the vector store. By default, Market0 will only include publicly available filings. However, users may also have access to analyst-level forecasts, providing them with additional insights when the response is generated.
OKTA FGA (Fine-Grained Authorization) is used to check which documents the user has access to based on their permissions.
How it works
- User Forecast Request: The user requests a forecast for a specific company, such as ZEKO.
- Document Retrieval: Market0 handles the request and employs a retriever to search its vector store for documents relevant to the requested information. It applies filters to ensure only the documents the user has access to are considered.
- Response Generation: Based on the retrieved documents, Market0 compiles a tailored response for the user. Depending on user's permissions the response could be based on analyst-level forecasts.
Explore the code
To implement this functionality, we use several helper functions and components, detailed below.
1// Get the db vector store
2const vectorStore = await getDocumentsVectorStore();
3
4// Create a retriever that filters the documents by symbol
5const retriever = vectorStore.asRetriever({
6 filter: { symbol },
7});
8
9// Create a Retriever Wrapper that filters the documents by user access
10const fgaRetriever = new FGARetriever({
11 retriever,
12 fgaClient,
13 buildQuery: (doc) => ({
14 user: `user:${claims.sub}`,
15 relation: RELATION.CAN_VIEW_DOCS,
16 object: `doc:${doc.metadata.id}`,
17 })
18});
19
20const docs = await fgaRetriever.invoke('forecast of ZEKO');
21console.dir(docs);