Computer Vision Engineer

## About the role We are looking for a Systems-First Computer Vision Engineer who specializes in the "Last Mile" of AI : taking a model and making it run continuously, reliably, and instantly on live video feeds. This role is not about training models in a notebook; it is about building the high-performance highways (Pipelines) that allow Vision AI to run in the real world. You will architect robust streaming architectures using GStreamer/RTSP and optimize inference for ultra-low latency on Edge and Cloud environments. ## Key Responsibilities - Architect Streaming Pipelines : Design and implement robust, real-time video ingestion pipelines handling multiple RTSP streams using tools like GStreamer, FFmpeg, and WebRTC. - Inference Integration : Take trained models from the ML team and integrate them into production pipelines. Your goal is to ensure the model runs stable, fast, and without memory leaks. - Latency Optimization : Obsess over milliseconds. Optimize data processing pipelines to ensure low-latency inference on both Edge devices (NVIDIA Jetson) and Cloud servers. - Fault Tolerance : Build "Crash-Proof" systems. Ensure that if a camera goes offline or a frame is dropped, the system recovers gracefully without manual intervention. - Framework Evolution : Maintain and evolve our proprietary vision framework by writing modular, reusable, and efficient Python code/libraries. - Performance Engineering : Diagnose bottlenecks in the system whether it's CPU, GPU, or Network and implement architectural fixes. ## Skills & Requirements - Video Engineering Mastery : Deep expertise in video streaming protocols (RTSP, WebRTC, FastRTC) and processing tools (FFmpeg, GStreamer). You know how to handle frame buffers, decoding, and encoding efficiently. - Core Vision Stack : extensive experience with OpenCV and Image Processing fundamentals. You understand geometry, color spaces, and pixel-level manipulation. - Production Python : Strong experience writing fault-tolerant, multi-threaded/async code. You understand how to manage resources in long-running processes. - Deployment Native : Hands-on experience with Docker is mandatory. You know how to containerize a complex vision application with all its dependencies. - Mathematical Foundation : Command over geometry and statistics for designing complex logic layers on top of model detections. - Mandatory Hardware Willingness : Computer vision and Python developers will not just write code. They must be willing to research and validate hardware (e.g., optimal camera placement, server sizing, safety analysis, cost-effectiveness). - Rejection Criteria : Reject candidates who demand 100% software work. Candidates seeking a 60-70% software split are a good fit. - Scope of Hardware Work : It does not mean building camera internals. It means understanding specs (RAM, processor, megapixels, focal length) to determine the right fit for the application, similar to checking specs before buying a smartphone. - Team Collaborati