Built a financial RAG using custom agents and open source LLMs for analyzing and summarizing the quarterly reports of 4 major organizations. This end to end application was submitted at the MSADS Hackathon 2024 at University of Chicago and secured an Honorable Mention.
Tech Stack :
Unstructured.io - Extracting images and tables
LlamaParse - Parsing text and tables
Phi3 Vision - Summarizing images into text format
Instruct -XL - Embeddings for RAG
Qwen2 - LLM for generation. Has a large context window of 32K
Col Bert - Reranker for beter retrieval
Llama 3 - Creating hypothetical queries
Streamlit - Deployment