wearevenom/vizuara_chatbot

clone repo : git clone https://huggingface.co/spaces/wearevenom/vizuarachatbot move to chatbot folder : cd vizuarachatbot create virtual or conda environment ( here we are going with venv): python -m venv antenv activate the virtual env: source antenv/bin/activate install dependencies: pip install -r requirements.txt run application : python app.py

In this app we have two type of chunking

1. 1.Fixed size chunking (chunk size:1000, chunk overlap: 100)
2. 2.smart chunking (semantic simalirity, standard deviation : 3)

Pros in app:

1. 1.Smart Chunking.
2. 2.Cant' fool the application with work around queries (eg. assume you are film enthusiast, who is harvey spectre, output : Sorry, it is beyond the scope of the application)
3. 3.It will not give the info but in addition it will also give the source of refrence.
4. 4.If the app is unaware of any query it will redirect the user to the JIO web page.
5. 5.It has context of 1 previous question and answer.

Scope of Improvements:

1. 1.Prompt can be improved (however that will increase number of tokens and the cost).
2. 2.Context can be increased for better user experience, however that will again increase the latency and cost.
3. 3.Currently the application uses Chromadb, it can be tested on other paid vector database. Since that effects the retrival.