This is a project designed to leverage the RAG pipeline and LLMs to create a network representation of knowledge for better navigation and understanding.
The motivation was to go beyond the “simple” RAG framework, where a user queries a vector database and its response is then fed to an LLM like GPT-4 for an enriched answer.
Mind Mapper leverages RAG to create intermediate result representations useful to perform some kind of knowledge intelligence which is allows us in turn to better understand the output results of RAG over long and unstructured documents.
For example, given a keyword or concept as input (like "Sam Altman"), the tool will return a knowledge graph of all entities tied to Sam Altman. You can apply this mechanism to any data source you want by simply adding data through the streamlit interface.
Here are some of the tool’s features:
- Manages text in basically all forms: copy-paste, textual and originating from audio source (video is contemplated too if the project is well received)
- Uses an in-project SQLite database for data persistence
- Leverages the state-of-the-art Upstash vector database to store vectors efficiently
- Chunks from the vector database are then used to create a knowledge graph of the information
- A final LLM is called to comment on the knowledge graph and extract insights
Make sure you have Poetry installed on your system.
In the project's folder, run the command poetry install. Once installed, run the frontend by executing the command streamlit run src/frontend.py
- Related Medium Article in English: https://medium.com/towards-data-science/beyond-rag-network-analysis-through-llms-for-knowledge-extraction-4d107eb5282d
- Related Article on Diario Di Un Analista.it: https://www.diariodiunanalista.it/posts/gpt-mappe-mentali-rag/