Using LLMs and Vector Databases for information retrieval over a website

A LangChain Powered app to query the contents of your website.

Head over to the home page in the web app and follow the instructions

• Crawls the url specified to generate a sitemap that is subsequently used to extract text data from the relevant webpages on that site
• Filter URL patterns using Regex
• Creates Vector DB (ChromaDB) of crawled pages using OpenAi embeddings
• Uses vector db:
  • To perform Similarity Search and fetch relevant links containing the contents similar to query
  • Use GPT-3.5 Turbo to simulate chatbot behaviour on the website's data

>>> git clone https://github.com/JayantTaneja/Ask-The_Site.git
>>> pip install streamlit
>>> cd Ask-The_Site
>>> pip install -r requirements.txt
>>> streamlit run Home.py

ChatGPT is a fine tuned GPT model capable of performing dialogue generation or 'chat' based on the knowledge/data stored in its weights. It is incapable of querying over an existing external knowledge base.

Using a vector db however, we can leverage the general knowledge of the LLM (GPT 3.5 turbo, in this case) to gain actionable results.

True, however, they rely on publically available information. Let's suppose you, at your company/organization have some private data that you do not want to expose for the purpose of preventing any data leaks. In such a case, embedding it in a vector db is a good alternative.

Secondly, With the framework LangChain, you have the option of using a locally hosted LLM like LLaMa(assuming you have necessary compute power). This web app aims to showcase the potential of such layouts.