At Harvested Robotics, we buildAI-powered laser-weeding systemsfor farms in India. We capture high-resolutionRGB + NIR imagery, train custom models, and deploy them on rugged edge compute to identify weeds with speed, precision, and safety.
We're looking for aComputer Vision Engineerwho lives at the intersection of algorithms, data, and real-time edge deployment someone obsessed with accuracy, latency, and pushing models into production on actual machines in the field.
What you'll do
- Design, train, and optimise CV/ML algorithms forweed detection, crop segmentation, and health classification.
- Build and refinedata pipelines: labeling, augmentation, dataset versioning, and quality control.
- Improveaccuracy, robustness, and FPSacross varied farm conditions (lighting, soil, rains, dust).
- Deploy and optimise models onedge hardware(Jetson, Qualcomm, x86, custom accelerators) with tight compute budgets.
- Implementclassical CV + deep learning(OpenCV, geometric transforms, segmentation/detection networks).
- Build and maintainreal-time inference pipelinesusingGStreamer, DeepStream, CUDA plugins, and custom kernels.
- Convert and optimise models usingTensorRT, ONNX Runtime, quantisation (INT8/FP16), and graph optimisation.
- Work with the perception team onNIR/RGB fusion, extrinsic calibration, and sensor synchronisation.
- Analyse field failures and iterate on modelsfast.
- Build internal tools forevaluation, visualization, and automated benchmarking.
What you need
- Strong grounding incomputer vision, machine learning, and image processing.
- Experience withPyTorch/TensorFlow, ONNX, TensorRT, or similar optimisation stacks.
- Comfort withC++/Python, CUDA, and real-time edge deployments.
- Hands-on experience withGStreamerand/orNVIDIA DeepStreampipelines.
- Solid understanding of model evaluation, metric design, class imbalance, and dataset bias.
- Ability to work with large, messy, real-world datasets.
- Experience withembedded GPU optimisation, Jetson NX/Orin deployments, or Qualcomm accelerators.
- Familiarity with robotics perception, ROS, or sensor integration.
Your models directly controlwhen, where, and how our robot fires a laser at weeds. Precision and reliability define product success, safety, and farmer trust. This is not a lab role your work runs on machines in actual farms every day.
