FactFlow Click here!
This web app allows users to input a research related query, which will be answered with an AI generated response. To reduce hallucination, this response is based on filtered and retrieved internet sources, and the text chunks from these sources that are found to be relevant to the query are included in the output. Users can also further explore on their own as the links to the supporting documents are provided.
The frontend of the application is developed using React, and it is hosted on Vercel. The main related files are QueryPage.js, App.js, Header.js and their corresponding CSS files. Some individual components in the query page have their own files as well.
Our backend is in Python, and we use Flask API to communicate between the backend and the frontend. The flask code can be found in flask_main.py. The backend is hosted on Heroku. We also use firebase interaction which is done in our flask_main files and tavily_main files, which is commented out for live demonstration purposes.
Once the flask recieves an HTTP request from the front end, it sends the query through our pipeline, which can be found in tavily_main.py.
Our pipeline uses Tavily API to find up to five relevant research papers/articles, which are stored in our firebase database.
Using the LlamaIndex framework, we embed each of the sources, and then retrieve five nodes (direct text chunks from the articles) that were found to be most relevant to the query. Then, using these nodes and the user's query, the query engine generates an appropriate response, stating clearly if not enough information was found to answer the question.