Retrieval Gap Analysis for Financial QA

Research project on measuring and reducing retrieval failures in RAG pipelines for long-document financial question answering.

This project studies retrieval failure modes in retrieval-augmented generation (RAG) systems for financial question answering. The goal is to identify where evidence is lost and improve end-to-end answer quality through retrieval-focused interventions.

Focus areas

  • Error decomposition for long-document retrieval
  • Retrieval quality diagnostics and ablation studies
  • Practical strategies to close the retrieval gap before generation

Ongoing outcomes

  • A reproducible evaluation pipeline for retrieval performance
  • Quantitative analysis of retrieval-to-generation failure transfer
  • A framework for testing retrieval improvements in realistic financial QA settings