Sr. Computer Vision Engineer

As Sr. Computer Vision Engineer, you'll own the models behind every detection we make, their accuracy, and their cost—working directly with the CPT O, alongside engineers who already cover the platform, camera ingest, and the app.

Responsibilities:

• The role is applied computer vision: making our detection models more accurate and cheaper on real lots and shipping those gains to production.

Parking Detection Pipeline (Primary Focus):

• Park-and-exit and enter-and-park detection; vehicle and person detection, tracking, and re-identification across multi-camera views.
• Accuracy on real lots: tuning thresholds, cutting false positives, and handling occlusion, night, weather, and crowded scenes.
• Working with the footage, we have compressed sub-stream video (around 900×500) plus periodic full-resolution stills.
• Gains come from better algorithms, not more bandwidth.

Model Training and Experiments:

• Building and labeling datasets from production footage; training and evaluating models against ground truth.
• Running tight experiments that end in a measured result and a ship/no-ship call for vehicle identity and plates.
• Optimizing license plate recognition usage (a third-party API today) and evaluating native alternatives.
• Vehicle identity matching for EV, reserved, and time-limited spot enforcement.

Cost and Scale:

• Lowering inference cost per property through GPU utilization, batching, and smarter triggering of expensive calls.
• Reducing dependence on any single camera vendor or third-party model API.

Condition and Event Detection:

• Native models for lot conditions (trucks, snow, hazards) that are handled by rules and LLM calls today.

Requirements:

• We need an applied computer vision engineer who ships. Our models run on real cameras and trigger real actions like towing and incident evidence. The job is measurable accuracy and cost gains in production, delivered in weekly increments, not open-ended research.
• 5-7+ years of professional experience, with at least 3 years on production computer vision or video systems.
• Strong in Python, hands-on with PyTorch and detection and tracking stacks (YOLO, DETR/RT-DETR, ByteTrack, DeepSORT, or similar).
• Re-identification and multi-camera tracking in real-world, uncontrolled environments.
• Strongly preferred: license plate recognition (ANPR / ALPR) pipelines in production detection, OCR, and reading plates from low-resolution camera footage.
• Video processing; ffmpeg, frame extraction, and working with low-resolution, compressed surveillance footage.