{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/92827fee7ea24c75b39f69aad02e708b\" frameborder=\"0\" width=\"1660\" height=\"1245\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1245,"width":1660,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1245,"thumbnail_width":1660,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/92827fee7ea24c75b39f69aad02e708b-71fe5898ff0c6e3b.gif","duration":354.561,"title":"Parikshiksha 2.0 Zero Hallucination RAC Pipeline 🚀","description":"In this Loom, I’m demonstrating Parikshiksha version 2.0, a production grade RAC pipeline for NCP focused on trust and zero hallucinations. I explain the hybrid retrieval engine, using PyMuPDF extraction, BM25 keywords, ChromaDB semantics, and Reciprocal Rank Fusion, then a cross encoder re ranker and grounded generation with chunk ID citations or a refusal. I show out of syllabus quantum entanglement triggering a refusal test. Our automated evaluator reports 87.5 percent correctness, 100 percent groundedness, faithfulness 1.0, across 32 complex questions. No action was requested from viewers."}