In focus
04
ChromaDB and Qdrant
Vector DB for Hungarian-language RAG and semantic search.
- Embedding pipeline for Hungarian-language documents (NAV records, SMB policies, terms of service).
- Vector indexes optimised for semantic search with Hungarian morphology.
- Hybrid filtering on metadata + similarity — category, date, ÁFA classification.
- RAG strategy with Hungarian-capable LLMs (Llama 3, Mistral) — relevant answers with source citation.
- Latency tuning for production inference, running on the GPU cluster.

