Job Summary:
Kinetic Automation is building a network of automated repair centers for modern vehicles. In this role, you will collaborate with engineers and researchers to develop and deploy vision models for tasks like semantic segmentation and object detection across 2D and 3D data.
Responsibilities:
โข Implement training loops, curate datasets, drive high-priority experiments, and partner with cross-functional teams to close feedback loops from edge cases
โข Collaborate on model development by implementing training loops, losses, augmentations, and evaluations using PyTorch
โข Keep current with the industry by summarizing relevant papers and PRs, and proposing small, testable improvements
โข Contribute to datasets by helping define labeling guidelines, curating splits, running quality checks, and maintaining data versioning
โข Run experiments to track metrics, perform ablations, write clear experiment notes, and present findings.
โข Provide production support by exporting models, writing basic inference code, adding tests, and assisting with performance profiling
โข Work cross-functionally, partnering with backend engineers on APIs, containers, and CI, and with ops/labeling teams on edge cases and feedback loops
Qualifications:
Required:
โข Deep ML / CV Fundamentals: You need hands-on experience training and evaluating deep models for segmentation and detection (PyTorch). You must understand how Transformer/LLM building blocks map to vision (ViT/DETR/Mask2Former) and have practical exposure to 2D/3D data, point clouds, and camera geometry
โข Curiosity & Strict Attention to Detail: You are obsessed with corner cases. You have a sharp eye for data anomalies, run rigorous ablations, keep meticulous experiment logs, and can clearly communicate trade-offs
โข AI-Empowered, Not AI-Dependent: We strongly encourage leveraging AI tools (Copilot, ChatGPT, Claude) to maximize your efficiency. However, you must 100% understand the underlying details of the code you ship. We are looking for strong independent thinkers and debuggers, not someone who simply passes along AI outputs without deep comprehension
โข Working knowledge of transformer and LLM building blocks applied to vision, including self-attention, positional encodings, tokenization, and mapping these ideas to vision models (e.g., ViT, DETR, Mask2Former)
โข Practical exposure to 3D/depth data, including familiarity with point clouds, camera geometry (intrinsics/extrinsics), basic calibration, and multi-view geometry
โข Proficiency in Python and the relevant tech stack: PyTorch, torchvision, Detectron2 or MMDetection/Segmentation, and Hugging Face Transformers
โข Strong communication skills with the ability to write tidy PRs, experiment logs, and short design notes to ensure reproducibility
Preferred:
โข Experience with Python services (FastAPI/Flask), Docker, and AWS services (S3, Batch/EC2, ECR) is preferred.
Company:
Kinetic Automation is pioneering effortless modern vehicle repair. Founded in 2021, the company is headquartered in Orange, USA, with a team of 51-200 employees. The company is currently Growth Stage.