Engineering the Intelligence Inside Willow's Smart Building AI Platform.

MLOps for smart buildings is the engineering discipline of deploying, monitoring, and maintaining machine learning models that process IoT telemetry - from HVAC, elevators, sensors, and power systems - at production scale across thousands of facilities. Without it, energy prediction and fault detection models degrade over time as building conditions change, data distributions shift, and new equipment is commissioned.

XFactr.ai built Willow's MLOps infrastructure from scratch: covering CI/CD model pipelines, model registry, automated retraining on data drift, shadow deployments, and A/B model testing - all via MLflow and Azure ML. Every model serving Willow's platform passes through this pipeline before reaching production.

Snowflake handles time-series IoT data at scale, supports multi-tenant data sharing for portfolio clients, integrates natively with Python ML frameworks via Snowpark, and provides the governance and security that enterprise real estate data demands. For Willow - ingesting over 10 million IoT telemetry points daily from globally distributed buildings - Snowflake's elastic compute model scales processing with actual usage rather than requiring fixed-peak infrastructure.

XFactr.ai uses Snowpark Python for feature engineering, Dynamic Tables for automated aggregation, Streams and Tasks for event-driven processing, and Cortex AI for LLM-powered analytics. This means the same data platform that stores raw sensor data also powers feature stores for model training and serves AI model outputs - without moving data between systems.

HVAC energy prediction AI trains on historical telemetry from individual units - compressor states, set-point temperatures, chilled water flow rates, ambient conditions, and occupancy patterns - to establish what a given unit should consume under specific conditions. XFactr.ai's models for Willow use XGBoost and LightGBM for general consumption prediction and LSTM networks for sequential time-series patterns.

In real time, the deployed model compares expected versus actual consumption. Deviations beyond a configurable threshold trigger anomaly alerts: the model identifies whether the issue is a set-point drift, equipment degradation, or a hard fault - providing root-cause context for facility technicians rather than a raw alarm. This moves maintenance from reactive breakdown response to planned condition-based intervention.

A building energy baseline is the expected energy consumption of a building under defined conditions - occupancy, weather, time of day, and operational schedule. Without an accurate baseline, energy savings cannot be measured, because you have no agreed reference point for what the building should have consumed.

AI-computed baselines are significantly more accurate than rule-based or regression benchmarks because they capture non-linear interactions between variables - for example, the combined effect of a heat wave, high occupancy, and an HVAC set-point change on energy demand. XFactr.ai's baseline models for Willow use feature engineering across weather API data, occupancy signals, historical telemetry, and operational calendar data to produce baselines that are measurable, auditable, and explainable to portfolio clients.

Operational AI for smart buildings refers to AI systems that continuously process live sensor data, HVAC telemetry, elevator usage, power meter readings, and occupancy patterns - and generate actionable operational intelligence in real time. This includes energy savings recommendations, predictive maintenance alerts, fault diagnostics with root-cause context, and sustainability reporting (ESG).

Willow's platform is purpose-built for Operational AI at scale - processing data from BMS, IoT sensors, CMMS, and third-party systems across global portfolios. XFactr.ai engineers the ML models, Snowflake data infrastructure, and MLOps platform that power Willow's intelligence layer. Without this engineering layer, the platform could not deliver consistent, auditable, production-grade AI across a growing portfolio.

XFactr.ai embeds directly into PropTech product teams as the AI and data science engineering capability - covering ML model development, MLOps platform engineering, Snowflake data architecture, IoT data pipeline design, and ongoing model monitoring.

Rather than delivering isolated AI projects, XFactr.ai builds sustained engineering capability that operates inside agile sprints, contributes to architecture decisions, and maintains production systems across a multi-year partnership. Engineers have deep context accumulated across years - they know Willow's data model, its platform requirements, and the operational reality of the buildings it serves. This continuity is what makes the AI reliable in production, not just impressive in a demo.

Three things: continuity, depth, and governance. XFactr.ai engineers have worked inside Willow's product since 2022 - they know the data, the models, the platform architecture, and the customers. That accumulated context means architectural decisions are made with full understanding rather than repeated onboarding.

The engagement operates at multiple cadences: daily engineering standups for sprint delivery, weekly product alignment for roadmap and priorities, monthly executive governance for programme-level progress, and quarterly strategic planning for the AI capability roadmap. This structure means technology and business strategy stay in lockstep rather than diverging between review cycles.

And crucially, XFactr.ai engages at CTO level - not just PM level. Architecture decisions, model design choices, and infrastructure trade-offs are made with the engineering leadership who actually build and maintain the systems.