Jason, James, and Verdify

Jason and James standing inside the Verdify greenhouse near the controller, sink, sensors, and north-wall equipment — Verdify started as a father-son greenhouse build in Longmont, Colorado. The AI layer came later, after the room was already real enough to hold the software accountable.

Verdify is a family public lab, not a SaaS landing page. The project began with a 367 sq ft greenhouse in Longmont, Colorado, then grew into a useful test case for local AI, deterministic edge control, and public evidence.

The core question is simple: can an AI planner improve a real greenhouse without becoming the thing that directly controls relays? Verdify answers that by publishing plans, telemetry, scorecards, costs, failures, lessons, and the exact AI-writable tunables that Iris (our OpenClaw AI agent) is allowed to change.

For serious questions, corrections, collaboration, build comparisons, or press, use the contact form.

367 sq ft

Physical Colorado greenhouse, not a simulated demo.

5 seconds

ESP32 firmware evaluates the real-time state machine locally.

Public proof

Plans, scorecards, telemetry, lessons, costs, and known limits are visible.

Local AI

Routine planning can run through OpenClaw and a local Gemma 4 26B A4B (MoE) vLLM route.

James Vallery

James is a computer science student at the University of Colorado Boulder and a builder with public work across Verdify, hackathon projects, and full-stack software experiments. His public footprint includes software projects, HackCU work, and the Verdify repository.

Computer science

CU Boulder student building toward practical software, AI tooling, and systems work.

Project builder

Public GitHub work includes Verdify and other software projects under the jrvallery handle.

Hackathon work

Token Gauge tracks AI API spend and was built at HackCU 12 with public Devpost and GitHub artifacts.

Public links:

Jason Vallery

Jason Vallery’s public work sits around AI infrastructure, cloud platforms, storage systems, product strategy, and community technology education. Professionally, he works in cloud product leadership at VAST Data. Locally, he is tied into Longmont’s technology community through Longmont NextWave and public AI education efforts.

Verdify is where that background meets a physical system that does not care if the architecture diagram is elegant. If the forecast shifts, the VPD band is wrong, or the controller pushes stale setpoints, the plants and scorecards say so.

AI infrastructure

Public work around VAST Data, cloud systems, and the data layer behind continuous AI.

Community AI

Founder of Longmont NextWave and speaker on practical local AI literacy.

Public builder

Open-source projects, public talks, and practical experiments around AI systems.

Public links:

The Project

The greenhouse now has climate probes, soil sensors, hydroponic monitoring, energy meters, weather feeds, cameras, and more than 170 Home Assistant entities. Those signals feed a public loop:

Grow

Crops define the target climate bands for temperature, humidity, VPD, light, and stress tolerance.

Plan

Iris runs through OpenClaw, reads greenhouse context, and writes bounded tactical intent.

Enforce

The ESP32 firmware owns relay decisions every 5 seconds and keeps safety local.

Measure

Telemetry, cost, climate stress, and equipment state become public evidence.

Learn

Scorecards and validated lessons feed future planning while noisy raw output stays inspectable.

The separation matters. Iris does not directly flip relays. It writes tactical intent. Firmware enforces safety. Telemetry records what happened. Scorecards judge the result. Lessons feed the next plan.

That is the public claim Verdify is making: every AI plan should become a falsifiable physical hypothesis.

Start Here

AI Greenhouse Control for the plain-English system overview.
Planning Loop for how Iris writes plans.
AI-Writable Tunables for the exact knobs the planner can set.
Safety Architecture for why the AI does not control relays.
Local Inference Setup for Cortex, vLLM, and Gemma 4 26B A4B (MoE).
Slack Operations for the human operations surface.
Live Evidence for public proof.
Lessons for what the system has learned.

Press And Media Notes

Verdify is a public, local-first AI greenhouse in Longmont, Colorado. Iris, an OpenClaw agent, writes bounded climate tactics from forecasts, crop bands, telemetry, lessons, and scorecards; an ESP32 controller keeps real-time relay safety local every 5 seconds.

Short description: Verdify is a 367 sq ft Colorado greenhouse used as a public testbed for AI-assisted climate planning. The project publishes plans, telemetry, scorecards, costs, failures, lessons, and known limits so claims can be checked against a physical system.

Editor notes:

Pronunciation: VER-duh-fy.
Location: Longmont, Boulder County, Colorado.
Project posture: personal public lab, not a SaaS launch or commercial greenhouse product.
Best one-line framing: AI writes tactics, ESP32 controls relays, public telemetry checks the result.
Contact route: use Contact Verdify and choose Press or Correction.

Media assets:

Social card1200 x 630 JPEG

Launch-ready card: local AI greenhouse, ESP32 relay boundary, public telemetry.

Greenhouse hero photoExterior greenhouse

The greenhouse lit at night in winter conditions.

ESP32 controller photoSafety boundary hardware

The controller that owns real-time relay decisions.

Cortex rack photoLocal inference hardware

The home-lab GPU rack used for routine local planning events.

Architecture diagramPublic-safe SVG

System flow from sensors and planning through dispatcher, ESP32 control, telemetry, and public evidence.

Useful links:

For corrections, use the contact form and include the page URL plus the specific claim or data point that needs review.

🌱 Verdify