Senior CV Engineer

Overview

STACK Construction Technologies is a prominent B2B SaaS company, with a robust foundation and a dedicated team of over 130 professionals. We are hiring a senior engineer to improve our computer vision system for construction documents. This role focuses on detection and segmentation quality, geometric accuracy, and production-grade pipelines that work reliably with messy real-world PDFs and drawings.

What You’ll Do

• Design, build, and improve end-to-end detection and segmentation pipelines for document images.
• Improve document ingestion for PDFs and other unstructured files, including parsing, rendering, and handling of multi-page and multi-view content.
• Increase model accuracy, boundary and geometric quality, and overall reliability of predictions for downstream use.
• Integrate modern detection and segmentation models into production workflows, and build the post-processing that turns model output into structured, usable geometry.
• Define evaluation metrics, investigate failure cases, and drive continuous quality improvements.
• Optimize latency, reliability, and cost across the inference and post-processing stack.
• Own training infrastructure, dataset curation, annotation quality, and continuous-improvement loops.
• Make architectural decisions and own system quality end to end.

Required Qualifications

• Experience building computer vision systems: detection, segmentation, or structured geometry extraction, used high volume in production.
• Experience working with messy, real-world image data or large unstructured visual datasets.
• Strong understanding of detection and segmentation tradeoffs, including model architecture choices, training data design, and post-processing.
• Ability to measure system performance with evaluation, testing, and production metrics.
• Ability to explain failure modes clearly and improve systems through debugging, dataset work, and iteration.
• Experience with multimodal models (vision-language models, document AI systems)
• Understanding of grounding — linking model outputs to source data or coordinates
• Backend engineering experience, including APIs, async processing, and scalable GPU services.

Preferred Qualifications

• Experience with polygon or mask post-processing, geometric regularization, or CAD-style structured output.
• Experience with layout-aware document processing, PDF vector extraction, or combining raster and vector signals.
• Background in document-heavy CV domains such as construction, real estate, medical imaging, geospatial, or similar workflows.
• Experience optimizing inference cost and latency at scale.
• Familiarity with open-source detection / segmentation ecosystems, training infrastructure, or model serving.

What Success Looks Like

• Accurate, geometrically correct predictions suitable for downstream measurement, and analysis use.
• Fast, reliable inference across large and messy real-world document sets.
• Clear quality metrics and a repeatable improvement loop.
• Systems that perform consistently under real-world production constraints.

Work Setup

• Fully remote.
• Time Zone/Location: We are considering candidates located in North, Central, or South America, with a preference for those operating in time zones ranging from PT (Pacific Time) through GMT+2.
• Async-friendly environment with a focus on ownership and deep work.

What This Role Is Not

• Not a model-training-only role. You'll own data, training, post-processing, and serving.
• Not a research-only role.
• Not a plug-and-play CV tools environment.