Skip to content

ChosunInsight

2026.03~06

Hybrid RAG chatbot

Question Academic query ROUTER Classify intent 3-way Structured answer Rule-based · calendar, etc. Hybrid search ChromaDB BM25 Dense + sparse → RRF Official site, live Separate path · always current LLM Generate answer With source docs Response

Problem

Academic information mixes structured, unstructured, and real-time data, so no single retrieval method could answer every question.

My part

Owned the RAG engine on a five-person industry-academia team, including three-way routing by question intent and the ingestion and index pipeline.

Result

Metadata for student ID, enforcement year, and whether a rule is current fixed the problem of outdated graduation rules being retrieved first. Web search restricted to the Chosun University domain (Jina, site:chosun.ac.kr) runs as its own path. QA tests passed went from 6 of 35 (17%) to 31 of 35 (89%), and the service launched on May 22, 2026. It is no longer running.
PythonFastAPILangChainChromaDBBM25Vue 3Docker