{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/543c1d7db0204c27a8cb65a87180f601\" frameborder=\"0\" width=\"1280\" height=\"960\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":960,"width":1280,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":960,"thumbnail_width":1280,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/543c1d7db0204c27a8cb65a87180f601-f598d13c78c1b099.gif","duration":300.148,"title":"Quralinkei Triple Stream Clinical RAG Pipeline","description":"Hi, I’m Earsh, this is Quralinkei, a clinical intelligence platform built on the moon stack with a triple stream rack pipeline powered by Llama 3.1. I walk through our architecture, retrieval pipeline, and a live demo using sources like PubMed, clinicaltrials.gov, and OpenAlex, with MongoDB Atlas for session and history. For retrieval we fetch about 80 PubMed candidates, 50 trials, and up to 210 OpenAlex candidates, score them with 50 percent relevance to Curie terms, 30 percent recency, and 20 percent source credibility, then pass the top eight diversely to the LLM. I demo inputs like Alzheimer’s, COVID diabetes, and lung cancer, plus follow ups that use conversation memory from the last three exchanges. No specific action was requested from viewers."}