Production-Ready Computer Vision ML Engineer

OvrScope, Inc.; Denver (CO); Hybrid; USD 120,000 - 160,000; Full time

Job summary

OvrScope, Inc. seeks a seasoned Computer Vision / Document Understanding engineer to own the data collection, model architecture, evaluation, and production inference under real latency and cost constraints. You will drive data labeling strategies, OCR integration, and layout-aware models from development to deployment.

The role emphasizes building in production, handling unlabeled data, and delivering end-to-end solutions for roofing imagery and estimate PDFs with scalable, efficient inference.

Qualifications

• 4+ years building computer vision or document-understanding systems in production.
• Strong Python and PyTorch, plus practical experience with OCR and layout-aware models.
• Have owned a model in production, including non-ideal parts.
• Comfort working without a labeled dataset handed to you.

Responsibilities

• Train and evaluate detection and document-understanding models on roofing imagery and estimate PDFs.
• Design labeling strategy and quality controls for a specialized domain with few public datasets.
• Build evaluation harnesses that reflect claim outcomes, not just benchmark scores.
• Deploy models behind FastAPI services with monitoring, versioning, and rollback.
• Optimize inference cost and latency for documents that can run hundreds of pages.
• Close the loop from production errors back into training data.

Skills

Python

PyTorch

OCR

Layout-aware models

Production deployment

Tools

FastAPI

Job description

OvrScope, Inc. seeks a seasoned Computer Vision / Document Understanding engineer to own the data collection, model architecture, evaluation, and production inference under real latency and cost constraints. You will drive data labeling strategies, OCR integration, and layout-aware models from development to deployment.

The role emphasizes building in production, handling unlabeled data, and delivering end-to-end solutions for roofing imagery and estimate PDFs with scalable, efficient inference.