Opsio is hiring a Visual Inspection Engineer to design, build and tune the camera-and-AI systems that find defects on our customers' production lines.

About the role

You own the imaging and the model: from choosing the camera, lens and light for a customer's part to training and deploying the network that catches its defects. Our customers are manufacturers. Every line brings a new material, a new defect and a new cycle time, and the gap between the software finds defects and “it finds your 0.3 mm scratch on brushed aluminium at 60 parts a minute” is where you work.

This is a hands-on, office-based engineering role. You will spend your days at the bench with sample parts, cameras and lighting rigs, and in front of a GPU training and evaluating models. You work closely with our Application Engineers, who take your setups to the customer's line, and you report to the Country Manager, Opsio.

What you will do

Imaging design

- Specify the full imaging chain for each application: camera (area or line scan, resolution, sensor size, interface), lens (focal length, working distance, depth of field, telecentric where needed), lighting (bright field, dark field, backlight, dome, structured, multispectral) and triggering.
- Build and test imaging rigs at the bench with customer sample parts; measure and document the resulting defect contrast, repeatability and cycle time.
- Run feasibility studies: capture image sets, prove whether a defect can be detected reliably, and write a clear go/no-go report with the hardware list and expected performance.

Model development

- Build classical image-processing pipelines in OpenCV (calibration, alignment, filtering, segmentation, blob and edge analysis) where they beat a neural network on speed or robustness.
- Design, train and evaluate deep learning models for defect detection, classification and segmentation: CNN backbones, YOLO-family detectors and anomaly-detection approaches for cases with few defect samples.
- Own the dataset: image collection plans, labelling guidelines, class balance, augmentation and versioning.
- Validate against the customer's acceptance criteria: false reject rate, escape rate and inference time per part.





Deployment and performance

- Optimise models for production GPUs and edge hardware (TensorRT, ONNX, quantisation, batching) so they hold the line's takt time.
- Set up and maintain the training and inference environment: GPU workstations and servers, CUDA, drivers, containers and model registry.
- Diagnose field issues escalated by Application Engineers, such as drift after a material change, lighting degradation or a new defect type, and ship the fix.

Team and product

- Create and maintain reference designs, imaging guidelines and model playbooks so setups are repeatable across customers.
- Feed what you learn back into the product: which defects recur, which hardware works, which workflows are too slow.
- Support pre-sales with demos and technical answers when a deal needs them.

What you bring We are looking for someone who knows both halves of the problem: the physics of getting a valuable image, and the models that read it.

Software and AI

1. OpenCV – strong, hands-on experience in Python and ideally C++: camera calibration, geometric transforms, filtering, thresholding, morphology, contour and blob analysis, template matching.
2. Deep learning – you have trained, evaluated and deployed models in PyTorch or TensorFlow on real, imbalanced industrial image data, not only benchmark datasets.
3. CNN – solid understanding of convolutional architectures for classification and segmentation (ResNet, EfficientNet, U-Net or similar), transfer learning, loss functions and how to read a confusion matrix against a customer's acceptance criteria.
4. YOLO – practical experience training and fine-tuning YOLO-family detectors (v5/v8/v11 or similar) for small-defect detection, including anchor and input-size choices,



augmentation and mAP versus false-reject trade-offs.

Hardware and imaging

1. Lighting – you can pick and justify an illumination technique for a defect type: bright field, dark field, diffuse dome, backlight, coaxial, line light, structured or multispectral, and you know how to kill reflections and ambient light.
2. Lenses – you can calculate field of view, working distance, resolution per pixel and depth of field, choose between entocentric and telecentric optics, and spot distortion or focus problems in an image.
3. Cameras – experience with industrial area-scan and line-scan cameras (GigE Vision, USB3 Vision, CoaXPress), sensor selection, exposure and gain, hardware triggering and encoder synchronisation on moving lines.
4. GPUs – comfortable setting up and using NVIDIA GPUs for training and inference: CUDA, drivers, Docker, mixed precision, and optimising with TensorRT or ONNX Runtime on workstation and edge devices such as Jetson.

Profile

- Bachelor's or Master's degree in Computer Science, Electronics, Electrical, Mechatronics or a related field.
- 3–6 years of relevant experience, with at least one visual inspection or machine vision system delivered to a factory floor.
- Clear written and spoken English; you can explain a trade-off to a quality manager and write a feasibility report a customer can act on.
- Based in or willing to relocate to Bangalore and work from our office full time.

Good to have
- Anomaly detection and few-shot methods for lines where defect samples are rare (PatchCore, PaDiM, autoencoders).
- Segmentation and transformer-based vision models (Mask R-CNN, SAM, DETR, ViT).
- Experience with vision SDKs and tools such as Halcon, Cognex VisionPro, Pylon, Spinnaker or Vimba.
- 3D vision: laser profilers, structured light or stereo for height and surface defects.
- Edge deployment on NVIDIA Jetson or industrial PCs, and PLC or fieldbus integration (Modbus, Profinet, EtherNet/IP, digital I/O).
- Experience with MLOps basics: experiment tracking, dataset versioning and automated retraining.
- Exposure to automotive, electronics, pharma, food or packaging quality standards

📌 Visual Inspection Engineer (Bengaluru)
🏢 Opsio
📍 Bengaluru