Hybrid Glossary-LLM System Cuts Translation Lookup Latency to Under 500ms

LectuLibre built a hybrid system combining a preprocessed bilingual glossary with an LLM fallback to deliver instant word explanations in translated books. The glossary handles common terms instantly, while DeepSeek covers rare words with caching. This approach reduced response time from 2-5 seconds to under 500ms, balancing speed, context, and cost.