ATLAS - by code for kaya

Inspiration

Massive construction projects rarely fail because a single truck is late; they collapse because thousands of invisible, interdependent commitments silently break. Traditional project management software operates like an autopsy report—it only shows you the damage after the project is already over budget and behind schedule. We were inspired to build Project ATLAS to replace reactive firefighting with proactive control, turning chaotic logistics into a predictable science.

What it does

Project ATLAS introduces an AI-native Promise Intelligence Engine that transforms every deadline, material delivery, and site inspection into a living, measurable object mapped onto a dynamic Promise Graph.

When specialized AI agents detect that a commitment is at risk (e.g., a critical steel delivery's confidence rating drops), ATLAS jumps into action. Instead of generic alerts, our Cascade Engine instantly calculates the precise downstream ripple effects, such as idle labor costs and financial penalties. Simultaneously, our Counterfactual Simulator tests thousands of alternative recovery paths in milliseconds to recommend the single most cost-effective action to keep the project on track.

How we built it

We engineered ATLAS as a multi-agent system built for the Kaya AI hackathon.

The Graph Layer: A dynamic graph database structure that maps tasks not as static timeline bars, but as interdependent "promises" with fluctuating confidence scores.
The Engines: Built using specialized LLM agents and predictive algorithms to drive the Cascade Engine (for impact evaluation) and the Counterfactual Simulator (for generating optimal path recommendations).
The Interface: A clean, intuitive dashboard designed to give site managers immediate clarity on what actions they need to take today, rather than what went wrong yesterday.

Challenges we ran into

The biggest hurdle was building the Counterfactual Simulator to model real-world chaos accurately. Simulating thousands of alternative permutations for highly interconnected dependencies in milliseconds required aggressive optimization of our agent orchestration layer. We had to ensure the AI didn't just find any path forward, but specifically optimized for the lowest financial impact while keeping downstream labor utilized.

Accomplishments that we're proud of

We successfully moved AI beyond the realm of simple "summarization chatbots" into a true operational intelligence system. We are incredibly proud of how our engine handles multi-layered cascading failures, offering concrete, actionable solutions to complex industrial bottlenecks in a fraction of a second.

What we learned

We learned that in heavy industry, data visualization isn't enough; predictability and actionable alternatives are everything. True project intelligence doesn't just mean flagging a problem sooner—it means presenting the fix before the problem even hits the ground.

What's next for ATLAS - by code for kaya

We plan to scale ATLAS by integrating deep historical telemetry from past construction projects to train our confidence scoring models further. Additionally, we want to integrate real-world IoT sensor data and weather forecasting APIs directly into the Promise Graph to automate risk detection even before a supplier reports a delay.