Building agentic video search that picks its own retrieval strategy
- Source
- x.com
- Date
Search videos as easily as you search text. voyage-multimodal-3.5 natively embeds text, images, AND videos. Our latest tutorial shows you how to build an agentic video search system using Voyage AI and @MongoDB! Link below 👇
💡Consider searching through cooking videos to find key moments: querying “onions caramelizing" requires visual understanding of color and texture, while “Gordon Ramsay's scrambled egg recipe” needs keyword matching on the chef’s name. No single search technique handles both query types well. This is where agentic search helps.
@MongoDB 💡In this tutorial, we implement an agentic search system that dynamically chooses between vector search and hybrid search on videos and their captions—adapting to each query type automatically.
@MongoDB Learn more in our tutorial: https://t.co/IZvJPhM1tP
Context
A video search for an action needs different evidence from a search for an exact name. Voyage AI's video-search tutorial addresses that difference by having a language model choose the search method. It splits videos into caption-aligned segments and uses voyage-multimodal-3.5 to represent each segment and its caption together as searchable numbers.
The router selects either similarity search over those representations or a that also matches words in captions. This lets visual descriptions guide one query while exact terminology influences another. The tutorial returns matching video segments with timestamps, making the result a route to the relevant moment rather than merely a whole video.
- multimodal — A model that works with more than text — reading images, audio, or video, and sometimes generating them too.
- hybrid search — Combining keyword and semantic search in one ranked result, so exact terms and paraphrases both find what they should.
- embedding — A list of numbers representing a piece of text's meaning, so that similar meanings end up numerically close and can be searched.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
Checking sign-in…
Loading comments…