GridLock Atlas

Overlaps
Example Coordinates
Top Lead
Close-up 3D Model
Simulation

Inspiration

Neighboring utilities plan new transmission lines years ahead, but each publishes its plan in a separate document. Along the Savannah River, Dominion Energy SC and Georgia Power can schedule rebuilds a few miles apart in the same years without anyone noticing. We wanted a tool that puts those plans side by side, one a planner could trust, so every fact comes from a public document with a quote and page number.

What it does

Atlas maps two utilities' public construction plans and flags project pairs that are within 25 miles and overlap in time. It ranks them as review leads and shows the evidence behind each one: distance, schedules, any documented coordination, and where sources disagree. You can explore a pair in 3D, estimate the impact of coordinating, and export a cited brief for planners.

How we built it

We collected 103 public planning documents, parsed them with Python, and geocoded substations using OpenStreetMap. A deterministic TypeScript engine measures distances and compares schedules. The app is Next.js with Mapbox GL and three.js. GPT-5.5 re-read every plan page as an independent cross-check, but no AI output is used as fact.

Challenges we ran into

Plans change between editions, and utilities publish budgets and deadlines rather than real construction dates, so we had to label timing carefully instead of guessing. Similar substation names (two different "Goshen"s) and a state line running down the river made location matching harder than expected.

Accomplishments that we're proud of

Atlas exactly reproduces Sperry's worked example and flags just 149 of 7,929 pairs (a naive rule flags 4,631). Every one of its 1,786 cited quotes is automatically re-verified against the original document.

What we learned

Honesty is a design problem: "no coordination found" has to read as unknown, not uncoordinated. And a simple, readable interface matters more than an impressive one.