video1.0<iframe src="https://www.loom.com/embed/02ca5ddc2bd94152805ce8973787d1bb" frameborder="0" width="1920" height="1440" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>14401920Loomhttps://www.loom.com14401920https://cdn.loom.com/sessions/thumbnails/02ca5ddc2bd94152805ce8973787d1bb-8adf5e920fe56ae0.gif266.393I built a hybrid retrieval system with AI assisted ingestion and four zones, each with its own pipeline, plus a human approval loop. Zone A handles images and videos using a lightweight vision model to generate metadata and tags for review and approval, then ingests into the vector tool. Zone B stores important links by fetching content, generating an LLM summary, and saving it. Zone C handles structured data, like up to 10 plan items at a time, and unstructured FAQs and policies with different chunking strategies for better retrieval. For queries like an iPhone comparison chart, we filter by zone, cohort, and offer ID, run parallel semantic and keyword search, re rank, and return the winner.