Document Vector Database

Overview

The Document Vector Database Connected System enables Large Language Models (LLMs) to answer user submitted questions based on Appian Knowledge Center Documents. By uploading documents to this connected system, users can perform semantic searches to pinpoint the most pertinent content related to their questions. The Connected System also boasts Client APIs tailored for the AI Knowledge Assistant Component. This allows the AI Knowledge Assistant Component to deliver AI generated answers to user inquiries sourced from documents stored in the database, as well as general questions.

Key Features & Functionality

  1. Upload Document - Uploads and stores the documents and its vector in the database.
  2. List Documents - Provides us the list of documents uploaded in the database.
  3. Database Operations
    1. Delete Documents - Enables us to delete the documents that are uploaded in the database.
    2. Sync Documents - Updates the existing documents in the database with the latest version of the document available in the Appian Knowledge Center.
    3. Change Database Password - Changes Database password.
  4. Query Documents - Get relevant pieces of content from documents for the given prompt.
  5. Generate Response - Perform search in the given documents and generate ChatGPT response for the given prompt.
  6. Client APIs for AI Knowledge Assistant component for fetching document details, chat completions, document querying, and uploading new documents to the database.

Notes:

  • Download the AI Knowledge Assistant, a sophisticated chatbot designed to perform semantic searches across your documents and provide precise answers to your queries.
  • This plugin is not supported for HA environments.
Anonymous
Parents
  • Hi can I know where the data is getting stored here and while use if azure open ai connection i not getting response

  • The data is being stored in an embedded database represented as an Appian Document which is created when your click "Test Connection" in the connected system. This database will store chunks of document text along with their embeddings. The user's query will bring up the top most relevant chunks from the database.

    Feel free to post the error you are receiving from Azure and I can try and help you debug.

Comment
  • The data is being stored in an embedded database represented as an Appian Document which is created when your click "Test Connection" in the connected system. This database will store chunks of document text along with their embeddings. The user's query will bring up the top most relevant chunks from the database.

    Feel free to post the error you are receiving from Azure and I can try and help you debug.

Children
  • Hi  ,

    Kindly answer the below questions:

    • Is there any external vector DB involved here where we can see our document embeddings?
    • In Appian as soon as we test connection this generated a database file, inside that DB we have lots of DDL, statements, and doc metadata, some chunks - who is responsible for triggers of that SQL?
    • Once embedded doc is generated , whatever documents we upload to vector DB via integration, those embedding gets stored in that same document only right , there is no other table/document involved in which we do indexing against our prompts? 

    Kindly let me know if we have correct understanding.

    Thanks

  • Simply search for the name you gave the database in the connected system by using the Appian objects search. This will allow you to find the embedded database document and the constant pointing to that document. If you want to create a new database with new documents, simply make a new connected system and name the database something different. 

  • Can we see where are they stored physically or the only way is to query them?

  • All data is being stored in a native H2 database represented as an Appian Document. Embeddings are created using OpenAI/Azure, but all embeddings/text chunks are stored within the Appian environment. 

  • Can I know where the embedded database is there is it like appian  cloud db instance or just representing a vector DB like pinecone ,can we physically see that? As it is the main concern to build private Ai. Thank you.