ML/AI/Computer Vision engineer

The role We're looking for a data engineer to own the full pipeline from raw sensor data to trained, validated models. The project involves multi-modal sensing in a live urban environment — LiDAR, cameras, and edge compute — producing research-grade outputs for a government client. You'll handle data ingestion, labeling, model training, output delivery, and field ops support. What you'll do TRAINING DATA PIPELINE • Ingest and organize multi-modal sensor data from Ouster LiDAR (point clouds), video, and radar into a structured data lake • Build labeling pipelines for AV detection training — annotating vehicles by rooftop morphology, sensor cluster geometry, and visual fleet signature • Work with reference data (photos, 3D scans, video) to construct ground-truth datasets for model training and validation • Implement data augmentation strategies for edge cases: occlusion, low light, adverse weather, high-density traffic MODEL TRAINING & VALIDATION • Train and evaluate object detection and classification models for AV identification using 3D point cloud and image fusion • Train behavior detection models for objectively defined traffic events — turn maneuvers, lane changes, acceleration/deceleration profiles — with explicit event logic, thresholds, and accuracy metrics • Optimize models for edge deployment on constrained hardware (low power, no GPU cluster) • Define and report precision, recall, false positive/negative rates per detection class; maintain a validation log against ground truth DATA OUTPUT & DELIVERY • Design and maintain the output schema: object-level tracks, event records, trajectory data, and aggregated summaries in CSV, JSON, GeoJSON, and Parquet • Ensure all outputs are timestamped, spatially referenced, confidence-scored, and fully documented for downstream research use • Build batch delivery pipelines to the data hub; handle connectivity degradation gracefully with on-device buffering • Strip or anonymize all PII (faces, license plates) at the edge before any data leaves the field device INFRASTRUCTURE & OPS • Stand up and maintain the edge-to-cloud data pipeline: edge inference → local buffer → cellular/fiber uplink → cloud storage → data portal • Monitor data completeness, system uptime, and sensor health during the live pilot • Support commissioning and calibration at field deployment locations What we're looking for Required • Python (NumPy, Pandas, PyTorch or TensorFlow) • Experience with point cloud data (LiDAR, PCL, Open3D) • Object detection pipelines (YOLO, PointNet, or similar) • Data labeling and annotation workflows • ETL pipeline design and batch data delivery • Geospatial data formats (GeoJSON, shapefiles, spatial referencing) • AWS or equivalent cloud (S3, Lambda, or EC2) • Git, CI/CD basics Strong plus • Multi-modal sensor fusion (LiDAR + camera) • Edge model optimization (TensorRT, ONNX, quantization) • Experience with ITS, smart city, or AV-adjacent deployments • Computer vision for traffic or roadway monitoring • Privacy-preserving ML (on-device anonymization) • Ouster or Velodyne LiDAR SDK familiarity • Research data pipeline experience (academic or government)