Computer Vision Engineer

We are seeking a skilled engineer to develop and maintain the system responsible for reading and measuring technical documents with precision and reliability. This role involves working with a variety of file types—some with well-defined structures like vector paths, layers, and transforms, and others that consist solely of pixel data. You will interpret structured data when available and create models to extract meaningful information when it is not. All solutions must be optimized to run efficiently on CPU, using limited memory and adhering to strict time constraints.

Key Responsibilities

- Parse vector files at the primitive level, including paths, strokes, fills, transforms, and layers, to reconstruct what a human observer would see.
- Filter out overlays and annotations that should not be included in measurements.
- Handle complex geometric elements such as lines, centrelines, skeletons, contours, topology, and address broken or duplicate geometry.
- Match labels to corresponding objects and verify their consistency with the underlying geometry.
- Accurately manage coordinates and units across rotations, nested transforms, and multi-scale pages.
- Extract tables, legends, and title blocks, including those that are visually drawn rather than encoded.
- Develop OCR solutions tailored for challenging cases where standard OCR fails, such as stroke fonts, rotated text, and densely packed pages.
- Detect small, repetitive, and rotated symbols on flattened pages.
- Extract hand-added markup and classify line styles, weights, dashes, and hatching patterns.
- Generate training data through synthetic examples and programmatic labeling.
- Perform rigorous evaluation on held-out datasets, with a focus on CPU performance.
- Define success criteria and measure performance against real ground truth before implementing changes.
- Maintain existing functionality without regressions.
- Prioritize speed and memory efficiency as core engineering requirements.
- Build visual debugging tools to facilitate understanding of what is correct and what needs improvement.
- Write clean, maintainable Python code and deliver validated work directly to production.
- Clearly communicate decision-making processes, including revisions and course corrections.

Required Qualifications

- Strong experience with vector graphics, computer vision, and geometric processing.
- Proficiency in Python programming and delivering production-quality code.
- Solid understanding of OCR technologies and challenges.
- Ability to work within strict resource constraints, optimizing for CPU inference and limited memory.
- Experience in building and evaluating machine learning models with honest and data-driven approaches.
- Familiarity with parsing complex document structures and handling multi-scale, nested transformations.
- Commitment to rigorous testing and validation against real-world ground truth data.

Preferred Qualifications and Benefits

- Background in vector graphics, CAD, GIS, or floor plan interpretation.
- Experience using vision-language models (VLMs) as supplementary tools rather than primary data sources.
- Skills in optimizing models specifically for CPU inference.
- Knowledge of TypeScript or additional programming languages alongside Python.
- Industry experience in construction, estimating, engineering, or other fields where accuracy is critical and errors carry significant cost.
- Proven ability to build ground truth datasets from scratch.
- Collaborative work culture that values measuring problems before solving, rapid iteration, and open communication with domain practitioners.
- Emphasis on honesty about uncertainty, prioritizing “we don’t know” over incorrect results.
- Ownership mentality: identifying gaps in document coverage and proactively fixing them.

This role offers the opportunity to work on cutting-edge document analysis technology that transforms technical drawings into trustworthy, verifiable data. If you are passionate about precision, efficiency, and continuous improvement in a challenging technical environment, we encourage you to apply.