Skip to content

About

a deliberately challenging benchmark built around real financial filings, where retrieving the right evidence is just as important as generating the answer.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

Developed a full Retrieval-Augmented Generation (RAG) pipeline for financial question answering using FinanceBench — a deliberately challenging benchmark built around real financial filings, where retrieving the right evidence is just as important as generating the answer. Started with a naive LLM baseline, then built the RAG system end-to-end: Processed financial PDF filings Added document and page metadata Chunked and embedded the data Built a FAISS vector index Retrieved relevant evidence Generated grounded answers from context only What made this project especially valuable was the evaluation. Measured the system across: Correctness Faithfulness to retrieved context Retrieval hit-rate Also ran improvement experiments with prompts, retrieval depth, reranking, and chunk sizes. My biggest takeaway: In RAG, a strong generator is not enough. If retrieval misses the right evidence, the whole pipeline can fail.

About

a deliberately challenging benchmark built around real financial filings, where retrieving the right evidence is just as important as generating the answer.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages