Lead ML Engineer

ZeroRFI•San Francisco, CA

4d•$270,000 - $310,000

About The Position

We are seeking a Senior ML Engineer to own the machine learning function at Zero RFI — building, deploying, and continuously improving the models that power our construction intelligence platform. You will architect production ML systems, lead a growing team of engineers, and work directly at the intersection of deep learning, structured construction data, and the real-world workflows of owners, contractors, and project teams. This is not a research role. You will ship models into production, measure their impact on active construction programs, and iterate fast. You will also be a technical lead — mentoring engineers, setting ML standards, and collaborating with our Principal Engineer on system architecture and platform integration. This is a rare opportunity to apply state-of-the-art ML to one of the world's most data-rich and underserved industries.

Requirements

Bachelor's or Master's degree in Computer Science, AI/ML, Statistics, Computational Engineering, or a related field (or equivalent practical experience).
5–8 years of hands-on experience building and deploying ML models in production environments, with at least 2 years in a technical lead or senior individual contributor role.
Deep expertise with modern deep learning frameworks — PyTorch preferred — and strong proficiency in Python and scientific computing libraries (NumPy, SciPy, scikit-learn, Pandas).
Proven track record designing and shipping production ML pipelines in cloud environments (AWS SageMaker, Vertex AI, or Azure ML) with robust monitoring and retraining infrastructure.
Experience with NLP and LLM systems — fine-tuning, RAG architectures, prompt engineering at scale, and embedding-based retrieval (vector databases: Pinecone, Weaviate, Turbopuffer, or equivalent).
Strong foundation in computer vision — object detection, segmentation, or document understanding — using modern frameworks (YOLO, SAM, LayoutLM, or equivalent).
Experience with MLOps tooling: experiment tracking (W&B, MLflow), CI/CD for ML, containerization (Docker), and orchestration (Kubernetes or ECS).
Solid software engineering practices — clean code, code review, testing, version control — and the ability to collaborate fluidly with platform engineers.
Excellent communication skills: ability to explain model behavior, limitations, and tradeoffs to both technical teams and non-technical AEC stakeholders.

Nice To Haves

Experience with AEC data types: BIM/IFC schemas, construction schedules (P6, MS Project), RFI/submittal logs, cost databases, or CAD/drawing formats (DWG, PDF).
Familiarity with computational geometry, 3D scene understanding, or spatial data processing (Open3D, trimesh, PointNet++, or similar).
Experience with graph neural networks (PyTorch Geometric, DGL) for structured relational data — particularly useful for BIM element graphs and project dependency networks.
Background in time-series modeling for forecasting and anomaly detection in project performance data (schedule variance, cost burn, productivity metrics).
Knowledge of generative AI architectures (diffusion models, transformers, VAEs, GANs) and experience applying them to structured or domain-specific generation tasks.
Experience with reinforcement learning or multi-objective optimization for complex constraint satisfaction — relevant to construction sequencing and resource allocation.
Contributions to open-source ML projects or published work in relevant venues (NeurIPS, ICML, CVPR, or applied domain conferences).
Exposure to construction workflows, building codes, or the AEC project lifecycle — or genuine curiosity to learn it fast.

Responsibilities

Design, build, and deploy end-to-end ML pipelines — from data ingestion and feature engineering through model training, evaluation, and production serving — for AEC-specific use cases including document intelligence, schedule analytics, and cost prediction.
Architect scalable ML infrastructure using modern MLOps practices: experiment tracking (Weights & Biases, MLflow), model versioning, A/B testing frameworks, and automated retraining pipelines.
Build and maintain NLP/LLM pipelines for AEC document processing — RFI parsing and response generation, submittal log classification, contract risk extraction, and change order analysis.
Develop computer vision systems for construction drawing analysis, defect detection from site photography, and progress monitoring from reality capture data.
Deploy physics-informed models and time-series forecasting systems for project schedule prediction, cost escalation detection, and construction performance analytics.
Implement graph neural networks and geometric deep learning models for BIM/IFC data analysis, spatial coordination, and MEP system optimization.
Integrate ML models with industry-standard tools (Revit, Procore, Autodesk Construction Cloud) through custom APIs and data connectors, ensuring models consume and produce data in the formats construction teams actually use.
Define ML engineering standards: model evaluation frameworks, data versioning practices, testing strategies, and documentation requirements.
Drive ML strategy by evaluating emerging techniques and architectures — determining what to build, what to fine-tune, and what to buy — in collaboration with the CTO and Principal Engineer.
Collaborate with AEC domain experts (project managers, owners' reps, estimators) to translate field problems into well-scoped ML problems and validate outputs against ground truth.
Lead research initiatives where relevant and represent Zero RFI's technical perspective externally to establish credibility in the AEC ML space.
Partner with the Principal Engineer to integrate ML capabilities into core platform services — ensuring models operate as first-class components of the broader system architecture.
Own the ML layer of the data platform: feature stores, embedding infrastructure, vector search, and structured data pipelines that serve both real-time inference and batch analytics.
Champion responsible ML practices: bias evaluation, model transparency, and clear documentation of model limitations for non-technical stakeholders.