Imaging & Measurement Engineer

We scan large sealed loads with X-rays. From the scan we measure real physical numbers.

Then we check if those numbers match the declared description of what is inside

The work

You build the measurement pipeline:

- Convert raw images to log-attenuation. Resample to a fixed mm/px scale.

- Estimate mass from the image. Calibrate it against known reference weights.

- Compute a dual-energy material index.

- Build an expected range for each category. The archive is noisy, so use median/MAD.

- Split the load into regions. Match regions to the declared line items with optimal

  transport. Mass that nothing explains is the main output.

Deep learning comes later: SSL pretraining, anomaly detection on frozen embeddings, a

fine-tuned detector. First the physics must be correct.

Two things make this hard. The image shows the packaging more than the content, so a model

that predicts the declaration only learns to copy it. And our old negative labels are not

reliable, so we treat them as unlabelled and make verified data ourselves.

The first months are mostly Postgres, data ingest, and fuzzy record linkage. 

Requirements

- Good Python (NumPy, SciPy). Good SQL and data modelling.

- Classical image processing: morphology, connected components, projection profiles,

  frequency analysis, segmentation.

- Robust statistics: median/MAD, contamination, shrinkage, probability calibration.

- Careful experiments. This matters most. You split data by group and by time. You look for

  leakage yourself. You report variance over seeds. You choose a metric that fits the

  decision, not the metric from papers.

- Tests, CI, versioned artifacts, reproducible results.

Nice to have: 16-bit scientific or medical images, or sensor data where calibration and

drift matter; anomaly detection with few positives; some physics (attenuation, beam

hardening, scatter, detector artifacts); PyTorch fine-tuning.

This is not: LLM or generative AI work, a clean labelled dataset, benchmark chasing, or pure

research.

Useful backgrounds: medical imaging AI, industrial radiography/NDT, semiconductor

inspection, remote sensing, applied physics.