ssrajadh
sentrysearch
AIComputer VisionSemantic SearchVideo ProcessingEmbeddingsMachine Learning
View on GitHub 61
+340
// summary
SentrySearch is a tool that utilizes semantic search technology to quickly find content within video clips, supporting automatic localization and clipping of relevant videos via natural language queries. The system segments videos and embeds them into a vector space, achieving efficient cross-modal matching using the Google Gemini API or a local Qwen3-VL model. Users can easily index local video libraries and automatically generate short video clips containing key events based on search results.
// use cases
01
Quickly retrieve and clip video segments using natural language descriptions (e.g., 'truck stopping at a red light').
02
Support metadata overlay for Tesla dashcam data, displaying speed, location, and time information in exported clips.
03
Provide a fully localized search solution, enabling privacy-protected video analysis on GPU-supported devices without the need for API keys.