AEGIS — Predicting Procurement Risk Before It Delays Data Centers

The Problem

Modern data centers are no longer constrained by computing hardware—they're constrained by procurement.

Critical electrical equipment like transformers, switchgear, generators, and substations often have lead times stretching well beyond a year. A single inconsistent vendor update or unnoticed delay can silently push back a project's power-on date, resulting in millions of dollars in lost revenue. The frustrating part is that the warning signs usually exist weeks or even months before the delay happens. They are simply buried across vendor quotations, emails, RFIs, purchase orders, and procurement documents that nobody has enough time to cross-reference manually.

We wanted to build something that could surface those hidden risks before they became expensive realities.

Our Goal

Instead of creating another procurement dashboard, we wanted to build an intelligent procurement co-pilot.

Our objective was simple:

Read every procurement document, understand the project's critical supply chain, estimate the probability of missing the energization deadline, and recommend the best possible mitigation plan—with evidence. Rather than only tracking procurement, AEGIS actively helps project managers make better decisions.

How We Built It

AEGIS combines several modern AI techniques into one workflow.

Document Intelligence
- Procurement documents, emails, RFIs, quotations, and schedules are parsed using vision-language document extraction.
- Every extracted fact remains linked to its original source for transparency.
Contradiction Detection
- Information from multiple vendors is reconciled automatically.
- Conflicting lead times or delivery commitments are highlighted immediately.
Risk Prediction
- A forecasting model estimates the probability that procurement delays will impact the final power-on date.
- Instead of producing only a single prediction, conformal prediction provides calibrated confidence intervals.
Dependency Analysis
- Procurement dependencies are modeled as a graph.
- Delay propagation identifies which components actually affect project completion and which have scheduling flexibility.
Optimization
- When a risk is detected, an optimization engine evaluates alternative mitigation strategies such as expediting shipments, dual sourcing, or resequencing construction activities.
- The recommendation is mathematically optimized rather than generated through intuition alone.
Human Approval
- Every recommendation remains reviewable before execution.
- The workflow maintains a complete audit trail to support high-stakes enterprise environments.

Challenges We Faced

The biggest challenge was integrating multiple AI systems into one coherent workflow. Each individual component—document understanding, machine learning, graph analysis, optimization, and agent orchestration—already exists independently. The real engineering challenge was designing a pipeline where these systems cooperate while keeping every prediction explainable and every recommendation traceable back to supporting evidence. Another challenge was the lack of publicly available procurement datasets. To overcome this, we designed synthetic procurement scenarios based on publicly reported industry lead-time ranges while ensuring the system architecture remains adaptable to real enterprise data.

What We Learned Building AEGIS reinforced that AI should not simply generate answers—it should help people make trustworthy decisions.

We also learned that explainability matters just as much as prediction accuracy. Procurement professionals need to understand why a delay is likely and why a mitigation strategy is recommended before taking action.

Combining machine learning with optimization and human oversight creates systems that are significantly more practical than standalone predictive models.

Future Work

We plan to extend AEGIS by adding:

Live ERP and procurement system integrations
Real-time supplier performance monitoring
Multi-project portfolio optimization
Continuous learning from completed projects
Scenario simulation for strategic procurement planning Ultimately, we envision AEGIS becoming an intelligent procurement assistant that continuously monitors project health and proactively prevents delays before they impact construction schedules.

Why AEGIS?

Most procurement platforms help organizations track projects. AEGIS goes one step further. It predicts, explains, and prescribes—transforming procurement data into actionable decisions before delays occur.