CMU Pittsburgh QA System

Collaborated in a team of three to build an end-to-end retrieval-augmented generation (RAG) QA system with multi-granularity chunking. Designed a Small-to-Big hybrid retriever that fuses BM25 sentence-level retrieval with MiniLM-L6-v2 paragraph-level dense similarity via weighted score aggregation, achieving 75% accuracy under LLM-based evaluation.