HybridRAG-BN Framework Enhances Bangla KBQA Using Gemma-4-31B-Instruct
August 14, 2026
HybridRAG-BN combines BM25 and BGE-M3 hybrid retrieval with a LoRA-fine-tuned Gemma-4-31B-Instruct model for answer verification in Bangla. The system utilizes GGUF quantization and DuckDuckGo-assisted fallback retrieval to improve robustness in low-resource settings.
HOW THIS AFFECTS YOU
●
builderYou can implement this architecture to improve performance for low-resource language knowledge base QA.