Computer Vision Engineer

About Sakshi

Sakshi is Thinkneural's visual intelligence product a computer vision platform that enables businesses to understand behaviour, movement, and activity patterns through cameras and video feeds. From people flow and dwell time analysis to anomaly detection and retail shelf intelligence, Sakshi turns video into business insight.

Roles and Responsibilities

• Design, develop, and deploy computer vision models for real-world applications including object detection, person tracking, pose estimation, and activity recognition.
• Build and optimise deep learning pipelines for processing video streams and image data at scale in production environments.
• Collaborate with product and data teams to translate business problems into computer vision solutions with measurable outcomes.
• Train, fine-tune, and evaluate models using frameworks like PyTorch or TensorFlow on custom labelled datasets.
• Develop and maintain data annotation pipelines, model versioning, and experiment tracking workflows.
• Integrate vision models into backend APIs and edge deployment environments (NVIDIA Jetson, ONNX, TensorRT).
• Work with camera hardware teams to optimise model performance under real-world constraints lighting variation, occlusion, low resolution.
• Monitor model performance in production, identify drift, and implement retraining and evaluation loops.
• Participate in code reviews, architecture discussions, and sprint planning with the Sakshi engineering team.
• Stay current with the latest research in computer vision and evaluate applicability to Sakshi's product roadmap.

Must Have

• 2+ years of hands-on experience in computer vision or deep learning in a production or research environment.
• Strong proficiency in Python with PyTorch or TensorFlow model training, evaluation, and deployment.
• Practical experience with object detection and tracking frameworks YOLO, Faster R-CNN, ByteTrack, DeepSORT, or similar.
• Solid understanding of image processing fundamentals OpenCV, image augmentation, preprocessing pipelines.
• Experience deploying models to production, including model optimisation (ONNX, TensorRT, quantisation, pruning).
• Familiarity with cloud platforms (AWS / GCP) and containerisation (Docker) for ML workloads.
• Ability to work with real-world video data and handle practical challenges like occlusion, variable lighting, and camera distortion.
• Strong problem-solving skills and ability to move from research prototype to production-ready implementation.

Good to Have

• Experience with edge deployment on NVIDIA Jetson, Raspberry Pi, or similar embedded platforms.
• Knowledge of multi-camera systems, camera calibration, and 3D scene understanding.
• Familiarity with MLOps tooling — MLflow, DVC, Weights & Biases, or similar.
• Exposure to retail analytics, surveillance systems, or smart space intelligence use cases.
• Understanding of video streaming protocols (RTSP, WebRTC) and real-time inference pipelines.
• Experience with data annotation tools (Roboflow, Label Studio, CVAT) and managing large labelled datasets.
• Research background or publication in computer vision — a plus but not required.

The Person We Are Looking For

You are someone who has actually shipped a computer vision system — not just run a notebook. You understand the gap between a model that works on a benchmark and one that holds up in a shopping mall at 6pm under fluorescent lighting, with a partially blocked camera.

You are curious about the applied side of the field. You follow research, you know what's new, but you are equally comfortable reading an inference latency log and debugging a frame-drop issue in a video pipeline.

You work well with ambiguity. On the Sakshi team, problems are often defined as business questions — "why does footfall in Zone 3 drop after 4pm?" — and you help translate those into the right visual signal to capture and the right model to run.

If you care about building vision systems that actually change how people make decisions — not just demos — we want to hear from you.