ML Engineer (Computer Vision & Document Intelligence)

🚀 We’re Hiring: ML Engineer (Computer Vision & Document Intelligence) (Remote | Full‑Time | SaaS | AI‑Native Platform)

Company: FieldFlo, USA. Industry: Construction • Environmental Services • Demolition • SaaS • AI

FieldFlo is a fast‑growing, mobile‑first SaaS platform built for the construction, demolition, and environmental services industries. Teams across the U.S. rely on us daily for compliance, safety, time tracking, training, and seamless field operations. We’re backed by industry veterans and growing fast — this is your chance to directly influence a platform that keeps workers safe and projects running smoothly.

🌟 Role Summary:

This is a specialized, hands-on ML engineering role focused on building production computer vision and document understanding systems for the construction industry. You’ll design and ship AI pipelines that extract structured data from technical drawings — demolition plans, floor plans, architectural schematics — and turn them into actionable estimates and scope-of-work items inside FieldFlo.

The core challenge: demolition and abatement contractors receive architectural drawings that vary wildly across firms — different keynote systems, different legends, different symbol libraries, no enforced standard. Your job is to build systems that reliably detect, classify, and extract meaningful data from these documents, combining custom-trained computer vision models with LLM-based reasoning in a multi-stage pipeline.

This is not a research role. You’ll be shipping production systems that real estimators use daily to build bids. You’ll work closely with our product and engineering teams, and you’ll collaborate with an external AI partner (Steelhead) who has built a reference pipeline for manufacturing drawings that informs our architecture.

💡 What You’ll Work On

Build Computer Vision & Document Intelligence Pipelines (Primary Focus)

• Design, train, and deploy object detection models (YOLO-based) to identify annotation regions on construction drawings — keynote tags, dimension strings, room boundaries, title blocks, demolition legends, and notes.
• Build structured extraction models (Donut or equivalent document understanding models) that read cropped drawing regions and output structured fields — keynote codes, dimensions, room names, material specifications.
• Integrate vision pipeline outputs with LLM-based reasoning layers that cross-reference extracted data, resolve ambiguities, and produce estimate line items with per-field confidence scores and source traceability.
• Own the full pipeline from training data to production inference — annotation tooling, dataset management, model training on AWS (SageMaker or equivalent), evaluation harnesses, deployment, and monitoring.
• Build and maintain evaluation frameworks that measure extraction accuracy, track model regressions, and inform retraining decisions.

Document Processing & PDF Engineering

• Build robust PDF processing pipelines for construction documents — handling multi-page plans, mixed page sizes, scanned vs. digital-native drawings, and varying DPI/resolution.
• Implement page-level preprocessing: scale detection, orientation correction, legend extraction, and cross-page reference resolution (e.g., a legend on page 1 that applies to annotations on page 5).
• Design the annotation and labeling pipeline for training data — tooling, quality control, and dataset versioning for iterative model improvement.

Collaborate Across the AI Stack

• Work with the product engineering team to integrate pipeline outputs into FieldFlo’s estimation module — the extracted line items feed directly into the bidding workflow.
• Design the extraction output schema alongside the product team — source type, source reference (bounding box coordinates), per-field confidence scores — so vision pipeline results plug into the same review UX used for text-based extraction.
• Collaborate with our external AI partner (Steelhead) to adapt their reference architecture (YOLO + Donut + LLM) to the demolition/abatement domain.
• Contribute to LLM-based extraction pipelines for structured documents (Xactimate PDFs, Excel estimates) where computer vision is not required.Design, train, and deploy object detection models for construction drawings

🔧 Must‑Have Qualifications

✔ Computer Vision & ML

• Shipped computer vision models to production — not Kaggle competitions, not academic papers, not demos. Real users, real data, real feedback loops.
• Hands-on experience training and deploying object detection models (YOLO family, Faster R-CNN, or similar) — you’ve dealt with class imbalance, annotation quality issues, and production inference at scale.
• Experience with document understanding or OCR-adjacent models (Donut, LayoutLM, TrOCR, or similar) for extracting structured data from visual documents.
• Strong working knowledge of training pipelines — dataset preparation, augmentation strategies, hyperparameter tuning, evaluation metrics (mAP, precision/recall per class), and model versioning.
• Practical experience with GPU-accelerated training on cloud infrastructure (AWS SageMaker, GCP Vertex AI, or equivalent) — you know how to manage training jobs, select instance types, and optimize cost.
• Understanding of when to use custom CV models vs. LLM-based approaches vs. hybrid pipelines — you’ve made this decision in production and can articulate the tradeoffs.

✔ Engineering

• 3+ years of professional ML engineering experience with a track record of deploying models to production environments.
• Strong Python skills — PyTorch (preferred) or TensorFlow, plus proficiency with the ML ecosystem (NumPy, pandas, OpenCV, scikit-learn).
• Experience building inference pipelines that handle real-world document variability — scanned PDFs, mixed resolutions, noisy inputs.
• Comfortable working with AWS services — S3, SageMaker, Lambda, or Step Functions for ML workflows.
• Familiarity with Docker and containerized deployments for ML models.
• Strong written and verbal English communication — you’ll be collaborating across time zones daily.
• Self-directed and proactive — comfortable owning problems end-to-end in a remote team without hand holding

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

💼 What We Offer

• Full-time remote (with at least 2h overlap with US Mountain Time)
• B2B contract
• Paid national holidays (based on your country)
• Optional unpaid personal time off
• 3‑month probation
• High-impact role shaping the future of an AI-native SaaS platform

📣 Recruitment Process

1. Initial call with Recruiting Agency HC Solutions
2. Technical panel interview (Senior Engineer + Tech Lead)
3. Interview with AI Labs team
4. Final interview with VP Engineering & CTO