Inspiration

Every breakthrough in AI begins long before the first GPU is powered on, it begins with building the data center.

Hyperscale AI data centers are among the world's largest and most schedule-sensitive construction projects. A single delayed transformer, chiller, or generator can ripple through thousands of downstream activities, delaying commissioning, pushing labor into unavailable windows, breaching contractual SLAs, and ultimately postponing the compute infrastructure powering modern AI.

Yet the software managing these billion-dollar projects remains fragmented. Schedules live in Primavera P6, procurement data lives in SAP or Oracle ERP, engineering models exist inside BIM platforms, while RFIs, emails, contracts, and field updates are scattered across separate systems. Each platform understands only a fraction of the project, leaving project managers to manually connect the dots.

As infrastructure projects become larger and supply chains more volatile, the industry's biggest bottleneck is no longer the lack of data—it's the lack of reasoning across that data.

That observation inspired Aegis.

What we're proposing

Aegis is an Operational Intelligence Engine for mission-critical infrastructure that reasons across construction schedules, procurement, contracts, vendors, resources, and field updates to help project teams understand not only what changed, but why it matters and what should happen next.

Instead of treating projects as disconnected documents or isolated schedules, Aegis will represent them as an interconnected operational model where relationships between tasks, equipment, vendors, labor, locations, and contractual obligations become explicit.

When a disruption occurs, for example, a delayed shipment of medium-voltage transformers, the system will determine how that delay propagates through construction activities, evaluate the operational consequences, and propose practical recovery strategies such as resequencing work, reallocating resources, or switching to alternative approved suppliers.

Our goal is to move AI beyond answering questions toward supporting real operational decision-making.

How we plan to build it

Rather than relying on a language model to perform engineering calculations, we separate deterministic computation from AI reasoning.

Structured project information from schedules, procurement records, engineering models, and project documentation will be transformed into a unified operational graph. Deterministic Python algorithms will perform schedule calculations, dependency traversal, and optimization, while LLM-powered agents will interpret natural language updates, identify affected entities, coordinate optimization workflows, and communicate recommendations to project teams.

The proposed architecture combines:

Structured project ingestion from scheduling, procurement, and engineering data
A unified operational graph representing project dependencies
Deterministic Critical Path Method (CPM) calculations
Agentic reasoning to explore alternative execution strategies
An interactive dashboard visualizing disruption propagation and recommended recovery plans

This approach allows the AI to reason about engineering constraints without relying on probabilistic text generation for mathematical scheduling decisions.

Challenges we expect

Construction projects generate highly fragmented and inconsistent information. The same physical asset may appear under different names across procurement systems, schedules, engineering drawings, contracts, and site communications.

One of the biggest technical challenges will be reconciling these disconnected representations into a single operational model while maintaining confidence in every relationship created.

Another challenge is balancing autonomous reasoning with deterministic engineering constraints. Large language models excel at interpreting context but should not be trusted to perform schedule mathematics. Designing a system where AI coordinates optimization while deterministic algorithms validate every engineering decision is central to our proposed architecture.

Finally, we aim to demonstrate enterprise-scale reasoning within the constraints of a hackathon by constructing a realistic synthetic hyperscale data center dataset that captures the complexity of real-world supply chains without relying on proprietary project data.

What we hope to learn

Through Aegis, we want to explore a broader question:

Can AI become an operational reasoning layer for construction rather than another interface for retrieving information?

We hope to better understand how structured representations, deterministic optimization, and agentic AI can complement one another to support decisions in industries where accuracy, explainability, and operational constraints matter more than conversational ability.

What's next

If selected, we plan to develop a working prototype centered around a realistic hyperscale AI data center construction scenario.

The prototype will demonstrate how a single supply chain disruption can propagate through scheduling, procurement, labor availability, and contractual obligations, while autonomously evaluating multiple mitigation strategies before recommending the optimal course of action.

Beyond the hackathon, the same reasoning framework could extend beyond data centers to airports, semiconductor fabs, power infrastructure, bridges, tunnels, and other large-scale engineering projects. While the domain vocabulary changes, the underlying operational principles, dependencies, constraints, resources, and optimization, remain fundamentally the same.