nazish-qamar/document-qa-with-vector-database

A document Q&A app built with langchain that utilizes RetrievalQAWithSourcesChain chain to build RAG pattern-based chatbot for your custom document. The RetrievalQAWithSourcesChain ensures that the app responds with the list of sources.

- The UI app is built with Chainlit

- OpenAI embedding and chat models are used for the demo.

Prerequisite:

create ".env" file in the root directory with following key:

OPENAI_API_KEY = YOUR_API_KEY

To start the app run the following command in terminal:

chainlit run document_qa.py -w

Note: Before running Q&A app, one could also run the chroma_db_basics.py file to get a feel of the ChromaDB Vector by running the command:

chainlit run chroma_db_basics.py -w

Steps happening in the backend:

1. Creating chunks of the uploaded document.

2. Creating list of "Document" object with source as "metadata"

3. Creating embeddings of the "Document" object list using OpenAI Embedding model.

4. Creating Chroma vector database of embeddings

5. User query from the chainlit becomes the user prompt to the OpenAI chat model. Please note that if you want the output in a specific format then give the example format after your query. For example one case ask question as:

"Your question ### Your format ###"