Senior Computer Vision Engineer — Vision Foundation Models
PerPlant | Remote or Copenhagen
üåæ THE PROBLEM
Agronomic advisors know exactly what field conditions they need to detect—but building a custom detector takes months and a specialized ML team. We're collapsing that to days, without requiring advisors to touch model architecture or read a precision/recall table.
üîß WHAT WE'VE BUILT
‚Üí Foundation-model backbone trained on 4 years of geotagged field imagery (200k hectares, ~1M datapoints, 2cm/px)
‚Üí Data pipeline that automatically feeds field operations back into model improvement
‚Üí Infrastructure for few-shot fine-tuning and compressed deployment to NVIDIA Jetson hardware
üë§ WHAT WE NEED
A Senior CV Engineer to own few-shot adaptation of foundation models (Grounding DINO, DINOv3, CLIP) to novel agronomic classes, knowledge distillation to edge-deployable models (YOLO, RT-DETR), and the system-level reasoning between detection, scene-level classification, and semantic search. You'll report to our VP of ML and work directly with domain experts and field operators.
‚úì YOU SHOULD HAVE
‚Üí Hands-on with object detection (YOLO, DETR) and foundation models (DINO, CLIP, ViT)
‚Üí Shipped distillation/pruning/quantization to production hardware (Jetson, ONNX, TensorRT)
‚Üí Deep PyTorch, PyTorch Lightning, distributed training (DDP) experience
‚Üí Ability to explain model behavior to non-technical stakeholders without jargon (non-negotiable)
‚Üí Experience with one of: multi-object tracking, instance segmentation, embeddings, or compression
üéØ NICE TO HAVE
‚Üí Foundation-model adaptation (LoRA, prompt-tuning, lightweight fine-tuning)
‚Üí Edge device optimization (Jetson Orin, embedded Linux)
‚Üí Autonomous systems, robotics, or agricultural tech background
üß™ HOW WE HIRE
Real assessment, not whiteboard puzzles: few-shot detection on a novel class, distillation to edge hardware, system design, and code review (PyTorch Lightning). 6–8 hours async + 45 min live. You own exactly what you're being assessed on.
⏱️ WHY NOW
We've de-risked the foundation model and data flywheel. The constraint is now the team—building the inference-optimization layer that lets advisors iterate quickly. If you've wanted to work on vision problems millions of farmers face daily, this is the moment.
‚Üí Apply here or reach out to: [email protected]
‚Üí Learn more: www.perplant.com
-
ComputerVision #MachineLearning #FoundationModels #EdgeAI #Agriculture #Hiring
