This project is a Python application that leverages the power of GPT-4, the cutting-edge language model developed by OpenAI, to build a conversational AI Chatbot. The unique aspect of this chatbot is its ability to utilize a set of documents as a Custom Knowledge Base, allowing it to provide more context-aware responses.
The chatbot is served through a web interface powered by Gradio, which provides a user-friendly way to interact with the model.
The application uses the GPTListIndex and GPTSimpleVectorIndex classes from the gpt_index library to generate a searchable semantic index of the documents, stored as index.json.
When a user enters a prompt, the system:
- Converts the input prompt into a semantic vector representation, similar to the representations of the indexed documents.
- Identifies the document(s) in the index that are most semantically similar to the input prompt.
- Uses the GPT-4 model to generate a response based on the content of the most similar document(s).
Please note that the index.json is a local resource. The application does not send its entire content to the GPT-4 API during the conversation. It merely serves as a means of identifying relevant document context within the local environment.
Follow the instructions below to get the chatbot up and running:
- Ensure you have the latest version of pip installed:
pip3 install --upgrade pip
-
Place your PDF documents in the
docsdirectory. These will serve as your Custom Knowledge Base. -
Install the necessary Python libraries (MacOS):
pip3 install openai
pip3 install gpt_index==0.4.24
pip3 install langchain==0.0.118
pip3 install PyPDF2
pip3 install PyCryptodome
pip3 install gradio
- Once all the dependencies are installed, you can run the bot:
python3 bot.py
You should now be able to interact with your AI chatbot through the Gradio web interface!
docker build -t kb-predictor .
docker run -p 7860:7860 kb-predictor
docker run -e OPENAI_API_KEY=your_api_key -p 7860:7860 kb-predictor
This project is open source and available under the MIT License.
If you have any questions or would like to contribute, please open an issue or submit a pull request.