ml engineer for construction documents

FieldFlo is a mobile-first SaaS platform for the construction, demolition, and environmental services industries. Its platform supports compliance, safety, time tracking, training, and field operations, while its AI-native capabilities help contractors extract actionable estimates and scope-of-work items from technical drawings and documents.

• Design, train, and deploy YOLO-based object detection models for construction drawings;
• Build structured extraction models using Donut or equivalent document understanding models;
• Integrate computer vision outputs with LLM-based reasoning layers to resolve ambiguities and produce estimate line items with confidence scores and source traceability;
• Own the pipeline from training data to production inference, including annotation tooling, dataset management, cloud training, evaluation, deployment, and monitoring;
• Build and maintain evaluation frameworks to measure extraction accuracy, track regressions, and inform retraining decisions;
• Build PDF processing pipelines for multi-page construction documents, mixed page sizes, scanned drawings, digital-native drawings, and varying resolutions;
• Implement page-level preprocessing, including scale detection, orientation correction, legend extraction, and cross-page reference resolution;
• Design annotation and labeling pipelines, including tooling, quality control, and dataset versioning;
• Integrate pipeline outputs into FieldFlo’s estimation module and bidding workflow;
• Design the extraction output schema with the product team, including source types, bounding box references, and confidence scores;
• Adapt Steelhead’s YOLO, Donut, and LLM reference architecture to the demolition and abatement domain;
• Contribute to LLM-based extraction pipelines for Xactimate PDFs and Excel estimates.

• 3+ Years of professional ML engineering experience with a track record of deploying models to production environments;
• Shipped computer vision models to production for real users using real data and feedback loops;
• Hands-on experience training and deploying object detection models such as YOLO, Faster R-CNN, or similar;
• Experience with document understanding or OCR-adjacent models such as Donut, LayoutLM, TrOCR, or similar;
• Strong knowledge of dataset preparation, augmentation, hyperparameter tuning, evaluation metrics, and model versioning;
• Practical experience with GPU-accelerated training on AWS SageMaker, GCP Vertex AI, or equivalent;
• Ability to choose between custom computer vision models, LLM-based approaches, and hybrid pipelines, and explain production tradeoffs;
• Strong Python skills with PyTorch or TensorFlow and proficiency with NumPy, pandas, OpenCV, and scikit-learn;
• Experience building inference pipelines for scanned PDFs, mixed resolutions, and noisy inputs;
• Experience with AWS services such as S3, SageMaker, Lambda, or Step Functions for ML workflows;
• Familiarity with Docker and containerized ML model deployments;
• Strong written and verbal English communication;
• Self-directed and proactive, with the ability to own problems end-to-end in a remote team;
• Nice to have: Experience with architectural drawings, construction documents, or engineering schematics, familiarity with CVAT, Label Studio, Roboflow, or similar annotation tools, experience integrating LLMs into multi-stage AI pipelines, knowledge of PDF rendering, coordinate systems, and document SDKs, startup or SaaS experience.

• B2B contract;
• Paid national holidays based on the employee’s country;
• Optional unpaid personal time off;
• 3-Month probation;
• At least 2h overlap with US Mountain Time;
• High-impact role shaping an AI-native SaaS platform.