Verdify
What if a greenhouse could learn?
367 sq ft. Longmont, Colorado. 5,090 feet. 15% humidity. 95°F solar peaks. Five crops. One AI.
172 sensors feed a 7-mode climate controller that evaluates conditions every 5 seconds. Crop profiles define what each zone needs at each hour. An AI agent named Iris plans around the clock — adjusting 24 tunables at sunrise, sunset, and every transition in between. The system measures every outcome, scores every plan, and gets better.
221K+ climate readings
250 days operational
115 plans written
75 lessons learned
An ESP32 mode controller enforces those targets across fans, heaters, misters, and fog. At every solar milestone — sunrise, peak stress, tree shade, decline, sunset — Iris reads the forecast, the crop band, per-zone VPD stress, and what went wrong yesterday, then adjusts the 24 tunables that shape how the controller responds. The system measures every outcome and learns from every cycle.
Temperature
The golden glow behind the data is solar irradiance through the glazing. It is the primary driver of indoor temperature, more predictive than outdoor air temperature. On a 65°F day with 800 W/m² of solar, the greenhouse will hit 85°F. On the same 65°F day under clouds, it stays near 70°F. The AI plans around the solar forecast, not just the temperature forecast.
The green band is what the crops need, computed from diurnal profiles of the active crop mix. The solid green line is the 15-minute smoothed indoor temperature. Equipment dots at the bottom show the control response: fans engage when temperature exceeds the band, heaters fire when it drops below, and fog provides evaporative cooling at extremes.
Humidity (VPD)
VPD is the metric that drives plant transpiration. Each mister zone has its own crop-driven target. When east-zone lettuce is stressed at 1.0 kPa, the system mists to raise greenhouse humidity. When south-zone cannas are comfortable at 1.5 kPa, those misters stay off. Equipment bars at the bottom show mister activity by zone and fog. The AI adjusts pulse timing, water budgets, and vent coordination based on the forecast.
Light & DLI
The golden background is solar irradiance through the glazing. The green line is daily light integral (DLI) accumulation, which resets at midnight and climbs with the sun. The dashed line is the crop DLI target (the minimum the plants need). When natural light falls short, grow lights supplement. Yellow bars at the bottom show when the two grow light circuits are active. The dashed solar forecast shows what tomorrow’s light budget looks like.
Economics
Three utilities stacked: electricity (11.1 cents/kWh), natural gas (83 cents/therm), and water (0.48 cents/gal). Winter is gas-dominated. Spring shifts toward electric cooling and water for misting. The AI optimizes across all three: gas heating overnight costs less per BTU than electric, fog during solar production hours is nearly free, and mister water is the cheapest resource but often the most critical for VPD management.
Greenhouse power draw (watts) overlaid with solar irradiance. When the curves overlap, the greenhouse runs on solar. The Tesla Powerwall covers overnight demand. The entire software stack runs locally on a single VM.
The Greenhouse
Structure, zones, equipment, what we grow
Climate & Control
Heating, cooling, humidity, water, light, the ESP32
Intelligence
Architecture, planning, lessons, data pipeline
Evidence
Operations, dashboards, economics, daily plans
About
The story behind the system