Senior AI Architect - Computer Vision

Job Description Summary
The Senior AI Architect for Computer Vision leads the Computer Vision Center of Excellence for Wind Engineering, establishing shared frameworks, standards, and reusable practices that enable reliable visual intelligence solutions across engineering initiatives.
This role manages computer vision platforms, provisions project workspaces, access, and resources, and facilitates projects from onboarding through technical transfer. It guides imaging strategy, data acquisition and annotation, classical vision, deep learning, multimodal models, 2D/3D analysis, edge inference, and integration into engineering and field workflows.
Through technical guidance, team mentorship, and consistent validation and governance practices, the role ensures vision solutions become reliable engineering products with measurable inspection performance, traceable datasets, and sustainable deployment across manufacturing, laboratory, turbine, drone, robotic and service contexts.
1. Lead the Computer Vision Center of Excellence for Wind Engineering
• Lead the Computer Vision Center of Excellence, establishing shared frameworks for acquisition, preprocessing, training, evaluation, deployment, monitoring, and feedback across Wind Engineering initiatives.
• Establish reusable methods and approved patterns for detection, classification, segmentation, tracking, anomaly detection, optical character recognition, visual measurement, pose estimation, and 3D reconstruction.
• Guide consistent selection of classical computer vision, deep learning, vision-language models, and hybrid approaches based on accuracy, explainability, latency, data availability, and operational risk.
• Align teams on edge, cloud, and hybrid design choices for cameras, drones, robots, inspection devices, and engineering applications, promoting consistent practices and reuse across initiatives.
2. Create Visual Data, Annotation, and Dataset Governance Practices
• Define standards for dataset curation, labeling taxonomies, annotation instructions, quality control, disagreement resolution, lineage, versioning, and access control.
• Establish methods for handling class imbalance, scarce defects, weak labels, synthetic data, augmentation, and active learning.
• Require training, validation, and test splits that prevent leakage across assets, sites, time periods, and near-duplicate imagery.
• Define dataset documentation that records intended use, coverage, limitations, provenance, and known failure modes.
3. Define Model Development and Validation Standards
• Establish baseline methods, evaluation protocols, acceptance thresholds, and error-analysis practices for computer vision systems.
• Require metrics suited to the task, including precision, recall, F1, average precision, intersection-over-union, calibration, localization error, false-call rate, and missed-defect risk.
• Define validation across lighting, weather, backgrounds, camera configurations, product variants, sites, and operational conditions.
• Ensure model comparisons consider engineering consequences and workflow economics, not only aggregate benchmark performance.
4. Architect 2D, 3D, Multimodal, and Spatial AI Solutions
• Guide architectures that combine RGB, thermal, depth, LiDAR, ultrasound imagery, CAD geometry, inspection history, and engineering context where appropriate.
• Define patterns for photogrammetry, point-cloud processing, registration, dimensional analysis, change detection, and digital-twin alignment.
• Evaluate vision-language and multimodal foundation models for search, triage, assisted inspection, and reasoning while preserving task-specific validation.
• Ensure geometric accuracy, coordinate systems, calibration, and uncertainty are explicitly managed in measurement-oriented applications.
5. Design Human-in-the-Loop Inspection Workflows
• Define how models present findings, confidence, evidence, and recommended actions to inspectors and engineers.
• Establish review and escalation paths for uncertain, novel, safety-relevant, or high-consequence observations.
• Design feedback capture so accepted, rejected, and corrected predictions improve datasets and future model versions.
• Ensure AI augments accountable engineering decisions and does not obscure required inspection evidence.