Hierarchical RAG is a powerful conversational agent that integrates hierarchical topic modeling and the RAG (Retrieval-Augmented Generation) model for interactive, context-aware question answering. Notably, this project distinguishes itself by avoiding the use of vectorization, providing a unique approach to information retrieval. Additionally, it allows users to query any hierarchical data source, expanding the scope of information exploration.
- Download the spaCy NLP model for subject extraction:
python -m spacy download en_core_web_sm - Add your OpenAI API Key at the top of the main file
- Run the script
- Try out the example prompts below
- Experiment with different queries and topics to witness the system's ability to handle complex conversations
- Go crazy! (not too much)
- What is the size of the Milky Way galaxy?
- Compare the religious demographics of Atlanta and Toronto.
- How different are the student clubs of Georgia Tech and IIT Bombay?
- Compare the airports in Chicago, Berlin, and Copenhagen.
Currently, we fetch all data from Wikipedia. Adding any custom data source requires two APIs:
- fetch: to fetch data about a topic
- parse: to generate a topic hierarchy
More data sources, such as SEC filings, will be integrated in the future.