Perception Engineer

Job description

At OKSI, we believe in owning problems end to end — from first concept through final delivery. We see it through! If you're looking for a place where you can take real ownership, stretch your skills, a place to grow your career while owning meaningful, mission-driven work, this is where you belong. OKSI is seeking a Perception Engineer to architect and implement perception, detection, tracking, and pose estimation algorithms across EO/IR, multi-spectral imagery, and other sensing modalities. The successful candidate will drive the full pipeline — from dataset curation and model development through real-time embedded deployment, simulation-based and hardware-in-the-loop testing, and integration with flight and mission software — working closely with software, hardware, and mission teams. WHAT YOU'LL GET TO DO Algorithm & Model Development • Drive the architecture for image processing, object detection, tracking, and pose estimation pipelines for next-generation spacecraft and sensing platforms • Design and implement robust perception, detection, segmentation, and tracking algorithms (CNN/Transformer-based, e.g., Retina/FCOS/DETR/Mask2D/video models) in C++ and Python • Build and refine model architectures for EO/IR and multi-spectral imagery, driving foundation-scale datasets, training recipes, and robust generalization to long-tail and degraded conditions • Train and fine-tune models in PyTorch, with rigorous experiment tracking and reproducible training pipelines Data, Training & Evaluation • Build large, diverse datasets and labeling/QA pipelines, including augmentation strategies • Stand up training and evaluation pipelines (PR/ROC, mAP, latency, robustness suites), with continuous regression testing and model-update loops from field data • Perform statistical analysis to support algorithm characterization and verification Real-Time Deployment & Optimization • Optimize models for real-time embedded inference (quantization/pruning, TensorRT/ONNX Runtime), profiling CPU/GPU performance to meet tight throughput and latency targets on embedded (e.g., Jetson-class) hardware • Apply model compression techniques (INT8/FP16) and dive into CUDA backends for performance optimization and debugging • Combine vision outputs with auxiliary sensing (radar/LiDAR/RF) for confirm/deny, association, and track management using decision-level fusion Test, Verification & Integration • Test and verify algorithms via software-in-the-loop (SIL) and hardware-in-the-loop (HIL) simulation, using high-fidelity sensor models and real-world datasets • Create visualization, triage, and root-cause analysis tools for rapid insight from simulation, HIL, and flight logs • Define end-to-end test plans in coordination with hardware and flight teams • Instrument health metrics, drift detection, and graceful degradation logic; write clear tests and documentation mapped to performance requirements • Work with the software team to integrate algorithms into flight software