Sub-second, hallucination-resistant Hindi Voice RAG system built for Hacker House Goa 2026 (Task 2) using AI4Bharat MSMARCO-XI, Sarvam AI STT/TTS, multilingual-e5-base, and Groq LPU inference. #RAGInGoa - Team Zyzz [Nabil Thange]
When the corpus cannot support an answer, SvaraSetu declines. It does not replace missing evidence with a confident guess.
Voice-enabled RAG over ai4bharat/MSMARCO-XI - Sarvam STT, hybrid HNSW+BM25 retrieval, six chunking strategies, calibrated abstention and grounding guardrails. HH Goa 2026 Task 2. #RAGInGoa
Description: Voice-enabled multilingual RAG system built for HH Goa 2026. Transcribes speech via ElevenLabs, retrieves from MSMARCO-XI in ~53ms, and generates verified grounded answers.
Add a description, image, and links to the ragingoa topic page so that developers can more easily learn about it.
To associate your repository with the ragingoa topic, visit your repo's landing page and select "manage topics."