-
Nextjs -
Python -
Langchain
Nimbus RAG
RAG system for Nimbus (Nimbus is a fictional company), information retrieval using private data, vector store and graph structures. Built with Python, LangChain, ChromaDB, NetworkX, Next.js, and OpenAI models.
Functionality
-
Information Ingestion
The system loads and processes data from PDF, Excel, text, and Markdown files as its knowledge base.
-
Vectors and Graphs
The system retrieves information using vectors containing relevant data and graphs with meaningful structures to generate accurate responses.
-
Natural Language
The RAG system operates using natural language to ask questions about the uploaded documents.
Challenges
-
Chunk Creation
The first challenge was creating meaningful chunks when ingesting different file types, generating metadata, and optimizing the information.
-
Graph Construction
A significant challenge was designing graph structures to improve response quality. The system must be able to generate graphs that accurately represent nodes and information connections.
Current Issues
-
Graphs
The graph structure does not fully represent the data flow structure. As a result, when handling very general queries, the system may fail to identify the correct graphs to generate an answer.