AI Assistant

It's an attempt to learn how to take advantage of LLM and create a better product. First I am converting docs to vector database, Now whenever user is quering, first passing this query to vector db which is returning context using using similarity search. Pass question and context to LLM with prompt which will return answer only from the context provided. Also have support for chat history so particular session becomes relevant for the user.

Demo

1. Select the Data Source
   Choose the data source for querying.

   

2. Ask the Assistant
   Interact with the assistant by asking questions related to your selected data source.

   

Installation

To set up the AI Assistant, follow these steps:

1. Install required dependencies:

   pip install -r requirements.txt

2. Obtain an API key from Google and copy to .env file

   cp env_sample .env

3. Create the necessary directories:

   mkdir -p data/personal_notion metadata

4. Copy your documents into the data/personal_notion folder.
5. Convert documents to a vector database:

   python data.py

6. Launch the application:

   streamlit run app.py

Let's Understand Architecutre

When you run python data.py, it converts the documents into a vector database. Whenever you ask a question, the system retrieves relevant context from the vector database and sends it to the LLM using a structured prompt.

TechStack

1. LangChain
2. ChromaDB
3. StreamLit
4. Google Gemini

Potential Use Cases

Here are a few ways you can utilize LLMs for different projects:

1. Configuration Generator
   Automatically generate configurations based on input data.

2. NLP to SQL
   Convert natural language queries to SQL statements.

3. Document Summarizer
   Summarize lengthy documents into key points.

4. Teacher
   Use the assistant to answer educational questions and help with learning.

You can extend the LLM by integrating custom tools and functions, giving it access to specific data and functionalities based on your needs.

Star History