> Markdown version of [/jobs/ext/4097950-software-engineer-computer-vision](https://www.wearedevelopers.com/jobs/ext/4097950-software-engineer-computer-vision). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md).

---

# Software Engineer, Computer Vision

- **Company:** CircuitHub
- **Location:** Cambridge, United Kingdom
- **Salary:** £80,000 - £125,000
- **Contract:** Permanent contract
- **Skills:** Training Data, Computer Vision, C++ (Programming Language), Python (Programming Language), PostgreSQL, OpenCV, Software Engineering, Pytorch, Salesforce Lightning, Deep Learning, Agentic-AI, ONNX (Open Neural Network Exchange) Format, Glasgow Haskell Compiler
- **Published:** October 19, 2026
- **Apply:** https://www.indeed.com/viewjob?jk=85d4bf2f44f000b5

## About the Role

* You've shipped computer vision into production, beyond a notebook.
 * Strong software engineering in Python, C++ or a comparable language.
 * A mix of classical vision (geometry, calibration, template matching) and deep learning; you know when each is the right tool.
 * Experience with industrial or machine-vision cameras is a bonus.
 * You use coding agents to move faster and take full ownership of what they produce.
 * You like the physical world: lighting, lenses and parts rotated by 90°.

Stack

 * Vision: OpenCV, PyTorch and Lightning, ONNX Runtime on GPU, libtorch from Haskell
 * Cameras: Basler GigE cameras via pylon, high-speed cameras with GPUDirect capture
 * Services: Python and Haskell, PostgreSQL, Nix and NixOS

## Description

* Inspect boards. Improve automated optical inspection that finds missing, rotated and wrongly placed parts and bad solder, working alongside our AI agents.
 * Pick reliably. Train and deploy the models that find parts in bins and trays, centre them on the nozzle and catch a clumped or dropped pick.
 * Align precisely. Fiducial detection, calibration and image alignment for placements measured in fractions of a millimetre.
 * Build the data loop. Decide what to capture, build labelling tools, and turn operator verdicts into training data.
 * Go fast. Work with our high-speed cameras and GPU capture pipeline to move vision closer to real time.
 * Measure it. Track accuracy in production, find where errors come from, and prove every change against past results.