Agentic AI research-paper debate / learning tool using LangChain, OpenAI, RAG and Vector Databases. Pits two LLMs against each other and manages a debate on a hot topic in the user uploaded research paper.
- Receives uploaded research paper document, processes and stores it for RAG, and asks an LLM to provide debate topics with For and Against motions.
- Sends the For motion to one AI Agent and the Against motion to another AI agent.
- Manager AI Agent manages the debate, and stops it either when there's consensus or if a certain number of points hav been made by either side.
- Summarizes and concludes the debate.
- Processes uploaded Questions and Answers on Solar System.
- Splits data into chunks and prepares embeddings for the data using my locally deployed Llama 3.2 model and embeddings.
- Loads them into high-performance, production-ready Weaviate Vector database.
- Performs search to find relevant stored documents for a prompt and uses Hugging Chat LLM for inference.
- Processes uploaded Questions and Answers on Solar System.
- Splits data into chunks and prepares embeddings for the data using my locally deployed Llama 3.2 model and embeddings.
- Loads them into high-performance, Facebook AI Similarity Search Vector database.
- Performs search to find relevant stored documents for a prompt and uses Hugging Chat LLM for inference.
- Compares performance between Weaviate and FAISS.
Performs similarity search on Weaviate persisted documents to find relevant stored documents for a prompt and uses Hugging Chat LLM for inference.
Full Retrieval Augemented Generation process on persisted data.
- What are Vector Databases? https://www.youtube.com/watch?v=ebMkbWzFCnA
- LangChain for orchestrating all the steps: https://python.langchain.com/docs/tutorials/rag/#components
- Weaviate https://python.langchain.com/docs/integrations/vectorstores/weaviate/#environment-setup
- Weaviate Search https://weaviate.io/developers/weaviate/search/basics