AI · Hybrid RAG search
Truth Chatbot: hybrid retrieval over 7M+ records
A retrieval-augmented chatbot backend where a wrong answer isn't a minor bug, so grounding is everything.
- Where
- Celestial Wave Digital
- Scale
- 7M+ records
- Stack
- Rust · Qdrant · Meilisearch
01Context
Truth is an Islamic companion platform: a library, the Quran with tafsir, a mosque map, and an AI chatbot that answers questions grounded in the Quran and Hadith. I built its chatbot backend at Celestial Wave Digital.
02Challenge
Across 7M+ records of hadith, tafsir and related texts, pure vector search misses exact names and references, and pure keyword search misses meaning. The answer must be grounded in the right passages, fast.
03Architecture
A Rust backend for predictable performance at this volume.
Qdrant for semantic (vector) search and Meilisearch for typo-tolerant full-text search.
A hybrid retrieval layer combines both result sets before the context reaches the language model.
04Key decisions
Hybrid over pure vector search, so exact names, references and rare terms stay findable.
Purpose-built engines for each job instead of one database doing everything.
05Result
A RAG backend that serves grounded answers over 7M+ records.