🏗️ IronVision Forecaster

Turn today's site footage into tomorrow's game plan.

IronVision Forecaster watches construction site video, understands what got done, and tells you exactly what to do next — tasks, materials, crew, risks, all in one dashboard.

The Problem

Construction site managers lose nearly 20% of every workday to material transport delays caused by reactive planning. That's not a technology problem — it's a visibility problem.

Ironsite's smart hardhat cameras already capture hours of first-person footage every day. But right now, that footage mostly sits in storage. Every evening, a site manager has to manually review the day, figure out what got done, decide what comes next, and calculate what materials to stage for the morning.

That process takes 30–60 minutes per site, per day — and it's still guesswork.

We asked a simple question: what if the footage could plan tomorrow for you?

What We Built

IronVision Forecaster is a two-stage AI pipeline that connects what happened on site today to what should happen tomorrow.

Stage 1 — See: A vision model watches the day's footage and produces a structured report — what tasks were performed, what materials were used, how many crew were on site, and what issues came up.

Stage 2 — Plan: A reasoning model takes that report, compares it against the project blueprint, and generates a next-day plan — which tasks to prioritize, exactly how many blocks and bags of mortar to stage, who to assign where, and what risks to watch for.

The result is a single dashboard that a site manager can open every morning and know exactly what the day looks like.

Tech Stack

Layer	What	Why
Vision Model	Llama 4 Scout 17B via Groq	Fast, free, handles multi-image input well for construction scene understanding
Reasoning Model	Llama 3.3 70B via Groq	Strong at structured reasoning — maps observations to blueprint logic accurately
Frame Processing	OpenCV + NumPy	Extracts 20 frames, stitches into 2×2 grids to fit within API image limits
Dashboard	Streamlit + Plotly	Custom dark industrial theme with glass morphism UI and live video background
Backend	Python	Lightweight, fast to build, everything runs from a single pipeline command

Business Impact

What this gives a site manager:

Every morning, instead of flipping through notes and footage, they open one dashboard and see — here's what got done yesterday, here's what the team should tackle today, here's exactly how many blocks and bags of mortar to have on the floor, and here's what might go wrong.

That's not a nice-to-have. That's how you stop losing 20% of every workday to reactive planning.

Specifically, IronVision:

Eliminates 30–60 minutes of manual evening planning per site, per day
Prevents material shortages — the #1 cause of on-site idle time
Catches issues early — weather, delivery delays, crew gaps
Creates a data trail — daily AI-generated progress logs for every project
Scales across sites — same pipeline, different blueprints

What's Next

We built this in 36 hours. Here's where it goes from here:

Next-hour prediction — intra-day micro-forecasting from live camera feeds, not just end-of-day summaries
Automatic inventory triggers — when predicted material needs exceed on-site stock, auto-generate purchase orders
Multi-site dashboard — one view across all active projects for operations leaders
Weather-aware scheduling — pull forecast data and auto-adjust the next-day plan for rain, heat, or wind
Historical learning — train on past project data to improve accuracy over time
Crew optimization — recommend how to split workers across concurrent tasks for maximum throughput