Machine Learning Engineer — Aerial Image Classification

The role RAAD's vision models are the reason clients come back: asset detection, defect classification, change detection, and volumetrics run automatically on every dataset the moment it lands. Our detection stack is built on the YOLO family and custom classification heads, trained on one of the largest proprietary corpora of close-range aerial imagery in the industry — and it's growing every day. We're expanding the ML team to cover more asset classes, more industries, and more geographies. What you'll do Train, evaluate, and ship detection and classification models (YOLO-family, ViT-based classifiers, segmentation) for aerial inspection use cases: roofs, solar arrays, transmission hardware, pipelines, flare stacks. Own the full loop — dataset curation, augmentation strategy for aerial-specific challenges (scale variance, nadir vs. oblique, thermal/RGB fusion), training infrastructure, and deployment to both cluster and edge targets. Build evaluation harnesses that catch regressions before clients do, with per-class, per-region, and per-sensor breakdowns. Work in Python across the stack: PyTorch, ONNX/TensorRT export, and the tooling that keeps annotation, training, and deployment moving. Partner with the edge team to quantize and prune models for on-device inference. What you'll bring 4+ years building and shipping computer vision models to production, ideally detection/segmentation on overhead or industrial imagery. Deep, practical PyTorch experience and fluency in the modern detection literature and toolchain. Strong Python engineering habits — your training code is software, not a notebook graveyard. Experience with model optimization for deployment (TensorRT, ONNX Runtime, quantization) is a strong plus.