VLM Scene Logging
Every 10-second video segment is processed by a Vision Language Model (InternVL2.5 / Qwen2.5-VL) that generates a plain-English scene description and stores it as a searchable embedding.
AI That UnderstandsYour Cameras.
YugYog's AI engine combines event-based real-time detectors with Vision Language Model scene indexing - so your cameras don't just record, they reason.
Three Layers Of Intelligence On Every Camera Feed
Event Detection Layer
14 dispatched detector classes fire in real time - PPE, falls, fights, fire, smoke, intrusion, loitering, line crossing, and more. Sub-second inference on a single mid-range GPU.
Event Catalog
189 prebuiltzzzz event types with AND/OR composite operators. Build custom alert rules without writing code - combine detectors, zones, time windows, and severity thresholds.
Real-Time Vs. Forensic
Zero-friction deployment
Real-Time Detection
Embark on a journey into the ever-expanding world of artificial intelligence exploration. Discover the latest advancements and breakthroughs shaping the future.
Zero-friction deployment
Forensic Search
Every scene is indexed at ingest time. When you need to find a moment - by description, object, zone, or time - the search engine queries against scene-log embeddings and returns ranked clips. No manual scrubbing required.
Multiple Detection Types. 6 Industry Packs.
Agriculture Pack
Animal detection & counting, perimeter, vehicle monitoring
Retail Pack
Loitering, queue analytics, after-hours, shoplifting behaviour
Healthcare Pack
Fall detection, patient wandering, restricted area, workplace violence
Manufacturing Pack
PPE, forklift-pedestrian, restricted zone, spill, loading-dock
Warehouse Pack
Perimeter intrusion, after-hours, dock monitoring, cargo detection
Transit Pack
Passenger safety, vehicle detection, platform monitoring
Full AI Stack - Inside YourNetwork.
YugYog's edge agent runs the full AI stack inside your network - RTSP ingest, event detection, VLM scene logging - on a single mid-range GPU server. In air-gap mode, ML models run fully offline using our enforcement flag. No model downloads after deployment.
