RAG-Based User Content Summariser

UGC platforms are key sources for product recommendations and technical problem-solving, yet users encounter scattered, disorganized, and sometimes contradictory information.

Information overload on platforms like Reddit, Stack Overflow, and Rednote makes it difficult for users to extract reliable, concise answers without manually reviewing hundreds of posts.

Led the development of RBUCS, a system leveraging RAG with LLM that performs text classification to select optimal processing strategies — producing ranked recommendations with evidence for recommendation queries and comprehensive summaries for general queries. Our hybrid retrieval mechanism combines FAISS semantic search, BM25 text matching, and content quality signals (e.g., user votes/likes). Evaluated on 150 labeled query-document samples, achieving 100% Recall@5 and 89.44% Precision@5 on Rednote data.