Foundation
Concrete estimating done by hand, with measurements taken from photos and experience. We're building the math layer.
Executive Summary
Concrete contractors estimate jobs by looking at photos, visiting sites, and doing math on paper or in their heads. We are building a machine learning system trained on historical pour data combined with 3D rendering of site conditions to predict the gap between what the math says and what the pour actually requires.
Architecture
ML pipeline with 3D site rendering and historical pour data calibration
ML pipeline with 3D site rendering and historical pour data calibration
The Problem
Concrete contractors estimate jobs by looking at photos, visiting sites, and doing math on paper or in their heads. The experienced ones are accurate. But the knowledge lives entirely in their experience.
Tacit Knowledge
There is no system that captures why a particular foundation needs 30% more concrete than the square footage suggests. That knowledge lives in thirty years of watching concrete cure.
Trust Gap
Contractors trust their experience over any screen. The system has to earn credibility by being right on jobs they already know the answer to.
Calibration Data
Training requires real pour data from completed jobs - what was estimated vs. what was actually used. Getting contractors to share this data consistently is an ongoing challenge.
The interesting problem is not the ML. It is getting contractors to trust a number that comes from a screen instead of from experience.
What We Built
Machine learning on historical project data combined with 3D rendering of site conditions.
ML Estimation Model
- Training on historical pour data
- Gap analysis between predicted and actual
- Site condition factor modeling
3D Site Rendering
- Visual site condition mapping
- Terrain and grade visualization
- Foundation layout modeling
Currently in calibration - training the model against real pour data from completed jobs.
Execution
Data Collection
OngoingGathering historical pour data from contractor partners. Building the dataset of estimated vs. actual concrete usage across different site conditions and foundation types.
Data quality is the bottleneck, not model architecture. Contractors record data inconsistently, and normalizing across different measurement practices is the real engineering challenge.
Model Development
In progressTraining estimation models that account for site conditions, soil type, grade, and foundation complexity. Validating against known jobs where the actual pour data exists.
Interested in this kind of work?
Start a conversation →