Inspiration

Utilities handle hundreds of millions of smart meter readings every month, and according to the U.S. Department of Energy, over 5% of those contain gaps, spikes, or rollovers that require manual fixing. We realized that exception queues in systems like Oracle MDM are often processed one record at a time by human analysts, costing millions in back-office labor and delaying billing.

The idea for this project came from seeing how static Validation, Estimation, and Editing (VEE) rules handle the “easy 80%,” while the hard 20% still requires people. We wanted to create an AI agent that could take on that long-tail.

What it does

Energentic is an AI-powered MDM VEE Exception Auto-Resolver. We detect & optimize error-prone smart meter data. From missing intervals to spikes and rollovers, Energentic selects the best resolution strategy, applies high-confidence fixes, and audits every change w/ confidence score. In turn, we provide a regulator friendly trail and massively boost billing cycles while saving millions in processing.

How we built it

6 Step Project 1) Data simulation: Generated 100 location based minute meter readings with daily patterns, weather sensitivity, and noise. 2) Error injection: Added missing intervals, spikes, and rollovers to mimic exceptions. 3) Detection module: Built simple rule-based detectors to flag exceptions. 4) Resolution strategies: Implemented three fixers—rules, SARIMAX, neighbor averages. 5) Policy chooser: Selects the best fix method per case based on data quality and exception type. 6) Audit logging: Records before/after values, method, confidence, and constraints check.

Challenges we ran into

Primarily 2 challenges;

1) Creating synthetic data that looked and behaved like real utility meter streams. 2) Balancing automation vs. accuracy—only applying fixes when confident.

Accomplishments that we're proud of

We accomplished our 2 main goals; 1) Achieved 25–30% auto-resolution rate with >90% accuracy on injected errors. 2) Created a clean audit trail for every automated fix, meeting compliance requirements.

What we learned

We learned 2 key lessons (it seems everything came in 2's this hackathon :P) 1) How VEE rules work in MDM systems and where they fail on edge cases. 2) Techniques for anomaly detection and gap filling in time-series data.

What's next for Energentic

We have 3 main goals; 1) Develop a proper front-end to host our product. 2) Train a machine learning policy model using real-world human review feedback. 3) Identify funding sources and collect real data to optimize the Energentic model with.

Built With

  • git/github
  • matplotlib
  • oracle-mdm
  • pandas
  • python
  • sarimax
  • scikit
  • streamlit
Share this project:

Updates