generative-question-answering-chatbot

This is a generative quesiton answering chatbot

The following is it's workflow:

• The user gives its input (ideally a query)
• The relavant pre-indexed documents are fetched from pinecone
• The transformer model generates a long form answer based on the question and context.

Components used:

• The model used for text generation is vblagoje/bart_lfqa from huggingface.
• The model used for computing embeddings for indexing is flax-sentence-embeddings/all_datasets_v3_mpnet-base from huggingface.
• Pinecone is used for storing and searching through embeddings.

Installation Instructions:

Create a virtual environment (highly suggested)

python -m venv ven

Activate the enviroment

• Linux

source venv/bin/activate

• Windows

.\venv\Scripts\activate

Install the required libraries

pip install -r requirements.txt

Indexing:

• Create an account with pinecone and get a pinecone_api_key
• assign your pinecone_api_key to PINECONE_API_KEY in indexing.py
• Put all you documents in documents.txt. There should be an empty line between each document.
• Then run

python indexing.py

Running the bot:

• assign your pinecone_api_key to PINECONE_API_KEY in bot.py
• assign your telegram_bot_token to TELEGRAM_BOT_TOKEN in bot.py
• Then run

python bot.py

Note: The documents in my documents.txt are generated by ChatGPT. Hence, they might not be factual.