I Left MongoDB to Build the Multimodal Database
Summary
I ran search go-to-market at MongoDB. I sat with hundreds of teams shipping search. They could query rows, JSON, full-text, even vector search on documents. But the moment a customer dropped a video, image, PDF, or audio file — that data was invisible. Just sat in S3, untouchable.
About this video
I ran search go-to-market at MongoDB. I sat with hundreds of teams shipping search. They could query rows, JSON, full-text, even vector search on documents. But the moment a customer dropped a video, image, PDF, or audio file — that data was invisible. Just sat in S3, untouchable. Every team was building the same hack: pull the file, run it through 5 models, glue the outputs into Mongo, pray. Reinventing a multimodal pipeline from scratch — and it broke every time the data shape changed. There was no database for that. No platform. No category. So I left and built it. One platform. Any file in. Extractors run in parallel. Features stored. Query across all of it with a single API. mixpeek.com #MongoDB #MultimodalAI #VectorDatabase #AI #FounderStory #MachineLearning #SearchEngineering #Mixpeek