01Overview
Project description
About the Role:
You will lead the design, development, and deployment of real-time video analytics systems for physical security. You will combine classical computer vision pipelines with large language models (LLMs) and vision-language models (VLMs) to enable natural-language scene querying, automated incident narration, operator alert summarization, and open-vocabulary detectionwhile meeting the latency, uptime, and compliance demands of 24/7 security operations.
Key Responsibilities
1. Architect multi-camera NVIDIA DeepStream (GStreamer) pipelines for concurrent 4K/HD feeds with hardware-accelerated decode.
2. Train and fine-tune detection, re-identification, and tracking models for security scenarios (low light, occlusion, fisheye).
3. Integrate VLMs (e.g., Gemini Vision, Qwen-VL) for open-vocabulary queries, scene understanding, and anomaly explanation.
4. Build LLM-powered operator workflows: natural-language alert search, incident summarization, and automated report generation.
5. Optimize CV inference with TensorRT (INT8/FP16); manage LLM inference via Triton, vLLM, or equivalent serving for latency-sensitive paths.
6. Design hybrid pipelines: fast CV models for real-time detection plus asynchronous LLM/VLM reasoning and narrative generation.
7. Integrate with VMS platforms (Milestone, Genetec, Nx Witness) and ONVIF/RTSP camera APIs.
8. Own production readiness: monitoring, model drift detection, prompt/version management, and retraining pipelines.
Required Skills & Experience
1. NVIDIA & Inference Stack
2. LLM & VLM Integration
3. Computer Vision Domains (Surveillance)
4. Video Pipeline & Integration
5. Engineering & MLOps
Preferred / Nice-to-Have
1. Deployments in critical infrastructure (airports, transport hubs, data centers, stadiums).
2. Experience with NVIDIA Metropolis platform or partner ecosystem.
3. Fisheye dewarping and PTZ auto-tracking.
4. Night-vision / thermal camera fusion.
5. False-alarm suppression at scale (10k+ camera networks).
6. Privacy and compliance exposure (e.g., GDPR, CCPA, NDAA 889) in video systems.
7. Knowledge of IEC 62676 and/or PSIA standards. .
