Sr. Computer Vision/ML Scientist

Aicadium•San Diego, CA

26d•$140,000 - $175,000•Hybrid

About The Position

Aicadium's Applied Research team is seeking a senior computer vision and machine learning scientist to advance open-world visual understanding with limited labels across 2D recognition, multimodal inspection, and 3D/4D perception. The role applies anomaly detection, zero- and few-shot recognition, and semantic segmentation to systems used across multiple industries, building on 2D and 3D vision foundation models and vision-language models (VLMs). The scientist will also help shape the research roadmap, particularly in 3D/4D reconstruction, spatial perception, and world models. This is a hybrid role based in San Diego, where Aicadium's U.S. headquarters is located. The ideal candidate has deep hands-on experience in modern computer vision, has advanced models beyond the prototype stage, can own problems end-to-end, and can support the development of junior scientists. Aicadium is a global technology company delivering AI-powered industrial computer vision products to enterprises. Its international team includes data scientists, engineers, and business strategists, and works to operationalize AI in organizations. The role joins a growing team in an agile, casual, and collaborative startup environment, working across industries and use cases. The company is building a talent hub in San Diego while remaining open to working with people throughout the U.S.

Requirements

PhD or Master's degree in Computer Science, Electrical Engineering, Robotics, Machine Learning, or a related field, with 4+ years of hands-on computer vision/ML experience; doctoral research counts toward this requirement.
Expertise in at least two of the following: anomaly detection, zero- and few-shot (open-vocabulary) recognition, semantic segmentation, or multimodal (vision-language) learning.
Experience adapting vision foundation models (e.g., DINOv3, JEPA-family models, SAM series) and VLMs (e.g., the Qwen-VL series) to new domains.
Hands-on experience in 3D/4D reconstruction and spatial perception, including modern 3D foundation models such as VGGT, MASt3R, and Depth Anything 3, grounded in multi-view geometry and camera models.
Ability to independently define research problems and turn recent research into working prototypes, with rigorous experimental practice including strong baselines, ablations, and sound evaluation on real-world data, including rare-event and heavily imbalanced problems.
Strong proficiency in Python and PyTorch.
Solid foundation in applied mathematics, including linear algebra, geometry, probability, statistics, and optimization.
Clear written and verbal communication skills, including presenting research to technical and non-technical audiences and stakeholders.

Nice To Haves

Publications at top-tier venues such as CVPR, ICCV, ECCV, NeurIPS, or ICLR, or notable open-source contributions in related areas.
Experience bringing computer vision models from prototype to production or customer-facing deployment.
Experience generating synthetic training data with generative models or simulation, particularly for defects, anomalies, and rare or data-deficient scenarios and classes.
Domain experience in industrial inspection, manufacturing, aerospace, food processing, medical imaging, or other fields with similar rare-defect or visual verification problems.
Experience with world models, such as V-JEPA 2, NVIDIA Cosmos, or spatially grounded VLMs.
Experience with VLAs and transformer-based action planners.

Responsibilities

Own applied research on vision problems, including anomaly detection, zero- and few-shot recognition, and semantic segmentation, from problem framing and baselines through validated models.
Adapt and extend 2D and 3D vision foundation models and VLMs to industrial domains through fine-tuning, harness optimization, and distillation.
Develop multimodal vision-language methods for industrial inspection tasks.
Conduct applied research in 3D/4D reconstruction, spatial perception, world models, and physics-based simulations, from proof-of-concept prototypes to reusable internal capabilities.
Design data and evaluation strategies for low-label, long-tail settings, including annotation plans, active learning, and evaluation protocols that improve field performance.
Systematically analyze and improve deployed vision systems; collaborate with engineering and product teams to bring models to production and maintain and improve model quality after launch.
Provide technical guidance to junior scientists and communicate methods, results, and trade-offs to technical and non-technical stakeholders.