Description
We are seeking a Video Analytics Engineer with 3-5 years of hands-on experience to build, improve, and optimize computer vision and video processing solutions that convert raw video streams into actionable intelligence. This role is ideal for an engineer who is strong in implementation and deployment, while also being curious about model behavior, failure analysis, and practical experimentation to improve accuracy in real-world scenarios.

You will work at the intersection of deep learning, real-time video processing, and scalable backend systems, contributing to use cases across security, retail, traffic, and industrial environments. You will partner with senior engineers and research-oriented team members to adapt existing approaches, evaluate results, and help move promising ideas into reliable production systems.

Responsibilities
- Design and develop end-to-end video analytics pipelines covering ingestion, decoding, preprocessing, inference, and post-processing.
- Build, fine-tune, test, and deploy computer vision models for object detection, tracking, classification,



action recognition, and anomaly detection.
- Analyze model outputs, identify common failure cases, and support experiments to improve robustness under occlusion, crowd density, lighting variation, and camera diversity.
- Optimize inference performance on edge and cloud GPUs using TensorRT, ONNX Runtime, OpenVINO, or similar runtimes.
- Implement real-time video processing using GStreamer, FFmpeg, or DeepStream, with low-latency streaming over RTSP, WebRTC, or HLS.
- Support dataset curation, annotation review, data augmentation, and retraining workflows to improve model quality over time.
- Collaborate with product, QA, and senior technical teams to translate business requirements into deployable analytics features.
- Contribute to benchmarking, validation, and performance measurement using explicit metrics for accuracy, latency, and throughput.
- Engineer backend services and supporting

📌 Software Engr II (Pune)
🏢 Honeywell
📍 Pune