SiteSync

Inspiration

Materials get ordered, and then the chaos begins. We watched project teams on multi-million-dollar jobs constantly scrambling, asking the same frantic questions: What’s been approved? What’s delayed? Will the structural steel actually arrive before the cranes are rented? The realization hit us that construction doesn't suffer from a lack of data; it suffers from a lack of synthesis. The industry was trying to manage dynamic, global supply chains with static spreadsheets, and project managers were always reacting instead of predicting. We were inspired to build a system that didn't just passively track materials, but actively understood the supply chain—a true, intelligent brain for construction logistics.

What it does

SiteSync acts as a centralized nervous system connecting procurement, fabrication, and the active job site. It tracks materials from the moment a submittal is approved to the second it arrives on site. Instead of just showing a material's current location, SiteSync features an AI predictive engine that calculates the probability of schedule deviation. It translates complex supply chain variables into simple, actionable alerts on a real-time dashboard, categorizing every critical material as On Time, At Risk, or Delayed, while explaining exactly why that status was given.

How we built it

We engineered the backend to ingest highly disparate data streams—from factory floor updates and transit GPS to weather patterns and historical delivery logs.Because construction data is notoriously noisy and prone to extreme outliers (like a sudden port strike or a truck breakdown), standard predictive models fell short. To train our arrival-time prediction engine, we utilized a robust regression approach optimizing for the Huber loss. This allowed the AI to penalize small errors normally, while remaining mathematically resilient to massive, unpredictable delays.

Challenges we ran into

Data Silos and Messy Inputs: Ingesting clean API data is easy, but construction doesn't run on clean APIs. Extracting usable data from scanned PDF manifests, unstructured emails, and legacy ERP systems required us to build a resilient OCR and Natural Language Processing (NLP) pipeline before the data could even reach our predictive model.

The "Black Box" Trust Problem: Early in development, the AI would accurately predict a delay, but users wouldn't trust it because they couldn't see the reasoning. We had to pivot our UI strategy to build deep explainability into the dashboard, ensuring the system highlighted the specific underlying factors (e.g., "Customs hold at Port of Long Beach") driving the alert.

Accomplishments that we're proud of

We successfully bridged the gap between advanced machine learning and practical, boots-on-the-ground project management. Building an OCR pipeline that can accurately parse messy, handwritten vendor manifests and feed them directly into a machine-learning model was a massive technical hurdle that we cleared. More importantly, we built a tool that construction professionals actually want to use because it reduces their daily cognitive load.

What we learned

The most profound lesson we learned is that in construction tech, transparency is far more valuable than raw accuracy. A 95% accurate AI that cannot explain its reasoning will be abandoned in favor of a 70% accurate spreadsheet that a human fully controls. Earning the trust of the user is just as important as the underlying math.

What's next for SiteSync

Our immediate next step is building native integrations with major construction management software like Procore and Autodesk Build, ensuring SiteSync lives right where project managers already work. We also plan to expand the predictive model to include macro-level climate and geopolitical risk overlays, allowing teams to predict material shortages months before they actually impact the job site.