OnSite Copilot
Build 2026

OnSite Copilot

Real-time AI agents for rework-free construction — a multi-client copilot that lives where building teams already work.

OnSite Copilot is an AI platform that brings real-time, agentic assistance to the construction job site. The premise is simple: construction software has always served the back office, not the field. OnSite Copilot puts an AI copilot where teams already work — on their phones, on a wearable, and inside Procore — to catch safety hazards and schedule risks before they turn into rework.

As founder and CTO, I set the product direction and built the system end to end: a unified Python AI backend and a family of native clients across mobile, wearable, CAD, and the web.

The Problem

Months of ecosystem interviews kept surfacing one core pain: costly, avoidable rework. Errors compound when supervision is thin and projects grow more complex — and field workers have no AI tools built for their reality.

5–12%
Rework cost
of total project cost lost to on-site errors
1:15
Supervision ratio
one supervisor to ~15 workers leaves most work unobserved
80%
Cost overruns
average overrun as project complexity rises
~$30Bin avoidable rework lost annually in U.S. construction

Who It's For

The copilot targets the two roles who own delivery and quality on every project — focusing on the moments where on-time scheduling and implementation quality are won or lost.

Project manager scene — managing on-time delivery and sequencing
Project Manager
TimeBudgetQuality
Owns on-time delivery — scheduling and sequencing conflicts account for ~41% of the burden.
Superintendent scene — controlling on-site implementation quality
Superintendent
TimeQuality
Owns implementation quality — controlling on-site operations accounts for ~45% of the burden.

Existing tools only document progress after the fact. OnSite Copilot aims to prevent the bulk of these burdens before they compound.

The Product

OnSite Copilot is three integrated products on one AI backbone.

  • Mobile Safety Copilot: An on-device computer-vision app that turns any phone into a safety monitor — detecting PPE violations and site hazards in real time, with instant alerts and a hands-free "Hey Copilot" voice assistant. Offline-first, so it works where connectivity doesn't.
  • Procore Side Panel: An app embedded inside Procore that brings schedule intelligence to where project managers already work — one-click schedule validation, critical-path and risk analysis, and a natural-language interface for "what-if" scenarios.
  • AI Backend Platform: The engine behind both — construction-aware LLMs, a RAG knowledge base for company standards and specs, and a multi-tenant, API-first design.

A Wearable Built for the Field

The safety copilot runs on a phone, but it's designed for a hands-free, head-mounted camera — so the AI sees what the worker sees, and the worker keeps both hands on the work. The companion app surfaces guidance on any nearby tablet or device.

OnSite Copilot wearable hardhat with an integrated camera module
Hardhat-mounted camera feeds the safety copilot
OnSite Copilot app running on a tablet on site
On-Site App
AI step-by-step verification UI flagging a placement against the manual
Step Verification
3D photogrammetry reconstruction of a captured site
3D Capture

Vision-Driven Site Understanding

The copilot doesn't just see pixels — it turns a captured site into a structured 3D model it can reason about. Drag the handle to compare the real scene with the AI's vision-driven reconstruction.

AI vision-driven 3D reconstruction of the siteAI visionCaptured construction site sceneReal scene

How It Works

OnSite Copilot is, in effect, a "Grammarly for construction" — two complementary agentic workflows that catch errors before they compound.

Workflow 1 · Procedural Error Prevention
Schedule Planner AI Agent
Input
SOPs · building information model · construction phasing plans
Agent
reasons over plan & sequence
Output
schedule recommendations for sequencing conflicts
Workflow 2 · Operational Error Detection
Computer Vision + VLM Pipeline
Input
worker photos · GPS / time context
Agent
multi-modal, low-latency vision-language model
Output
real-time, context-aware error alerts

See It in Action

OnSite Copilot — product demo
OnSite Copilot — product demo

For the full story, here's the OnSite Copilot pitch from the E14 hackathon:

OnSite Copilot — E14 hackathon presentation
OnSite Copilot — E14 hackathon presentation

A Three-Layer Agentic System

The platform is organized as three layers — sensing at the edge, integration into existing tools, and a construction-aware reasoning engine at the core — all served by one unified Python/FastAPI backend.

Edge Intelligence & Sensing
Vocal copilot · camera for site information — delivered through the mobile app.
Workflow Integration
Plugins for construction management tools — procedure checking and schedule optimization inside Procore, AutoCAD, and Revit.
Reasoning Engine
Multi-modal AI with construction domain knowledge — a vision-language model over a construction knowledge base (RAG with ChromaDB, LLMs via OpenAI and Anthropic Claude), with Procore OAuth and a ~2,100-endpoint API proxy.