Technical practice
How would you design a model to detect and classify assets across satellite and drone imagery?
What to coverThe posting calls for detecting and classifying assets in both image sources. Suggested preparation: discuss differences in scale and viewpoint, how you would select a model approach, and how you would measure errors by asset class.
When would segmentation add value beyond object detection?
What to coverThe posting names expertise in both segmentation and detection. Suggested preparation: compare bounding boxes with pixel-level masks, including their annotation costs and usefulness for locating objects with complex shapes.
How would you evaluate tracking across successive images?
What to coverThe role description mentions tracking assets, although the listed model duties emphasize detection and classification. Suggested preparation: explain how missed detections and mistaken identities affect a track, and what evidence you would use to assess continuity.
How would you build a trustworthy training dataset from the listed image sources?
What to coverThe posting calls for cleaning, annotation and augmentation of satellite, drone and ground-based imagery. Suggested preparation: cover label consistency, representative coverage and train–test separation, then explain how you would check whether augmentation helps rather than distorts evaluation.
What preprocessing differences would you consider for EO, IR, SAR and hyperspectral imagery?
What to coverThe posting explicitly lists these satellite data types. Suggested preparation: explain why their measurements should not be treated as interchangeable, and how you would inspect image quality and preserve information relevant to each modality.
How would you make an imagery-processing pipeline scalable and reproducible?
What to coverThe stated work includes pipelines for large volumes of satellite, drone and ground-based imagery. Suggested preparation: describe stages from ingestion through preprocessing and training, with checks for data quality, traceability and processing bottlenecks.
How would you assess whether combining optical, thermal and radar imagery improves robustness?
What to coverThe posting calls for integrating these modalities to strengthen models. Suggested preparation: compare a combined approach with single-modality baselines, accounting for alignment, missing inputs and conditions in which fusion might hurt performance.
How would you balance detection accuracy against reliable real-time performance?
What to coverThe role calls for high accuracy, efficiency and real-time operation. Suggested preparation: discuss latency alongside detection errors, and explain how you would test a faster model for unacceptable losses in reliability.
How would you check the accuracy of a target’s reported position and visualization?
What to coverThe posting calls for accurately locating and visualizing identified targets. Suggested preparation: discuss the path from an image detection to a displayed position, sources of positional uncertainty and checks that a visualization does not imply more precision than the data supports.
How would you design a target-prioritization workflow that responds to operator input?
What to coverThe stated duty is to make prioritization adapt dynamically to operator inputs. Suggested preparation: explain how you would make changes traceable, surface uncertainty and keep the operator able to review the resulting order.
Behavioural practice
Tell me about a computer vision research, internship or personal project you carried through evaluation.
What to coverThe posting cites these forms of computer vision work. Suggested preparation: give the situation, your specific actions and measured result, including what the evaluation did or did not establish.
Describe a difficult technical problem you tackled from first principles.
What to coverFirst-principles problem solving is explicitly valued in the posting. Suggested preparation: explain the situation, how you tested underlying assumptions, what you changed and what result followed.
Describe a time you found a data-quality or annotation problem that affected a model.
What to coverThe role includes data cleaning and annotation. Suggested preparation: set out the situation, how you identified and addressed the issue, and the observed effect on training or evaluation.
Tell me about a time you worked with others to resolve a disagreement about a model or pipeline.
What to coverThe posting mentions collaboration and work spanning models and data pipelines. Suggested preparation: describe the situation, your contribution to a shared decision and its result, without attributing the outcome solely to yourself.
Describe a time you had to communicate a model limitation to someone using its outputs.
What to coverThe role connects vision outputs to situational awareness and operator-responsive workflows. Suggested preparation: give the situation, how you conveyed uncertainty or errors, and what changed as a result.