⚡ GridNeighbors

Find where neighboring power utilities are planning grid construction in the same place, at the same time, and what coordinating could save customers.

Built solo in 24 hours at ShellHacks 2026 (FIU), my first hackathon, for the Sperry Tech GridLock Challenge. Devpost submission · Demo video

About the live demo: during the event the app ran at gridneighbors.design on DigitalOcean App Platform, with MongoDB Atlas for shared data. That hosted version was shut down after judging. You can still run it locally in about two minutes; the processed data is included in the repo.

---

The problem

Utilities plan transmission construction (new lines, rebuilt lines, substation upgrades) years in advance, but neighboring utilities plan separately, often across state lines. Two crews can end up rebuilding lines a few miles and a few months apart, each paying to mobilize its own crews, equipment, laydown yards and right-of-way. Those costs go into each utility's rate base, which customers pay back through their electric bills.

FERC issued Order No. 1920 in 2024 because siloed, piecemeal planning makes consumers pay more than necessary. GridNeighbors makes coordination opportunities visible at the level of individual projects.

What it does

GridNeighbors compares Dominion Energy South Carolina (DESC) with Georgia's 10-Year Transmission Plan (Georgia Power and its planning partners), neighbors across the Savannah River.

• Parses the utilities' own public filings: 252 planned projects (44 DESC, 208 Georgia) with in-service dates, plus cost and rate-base year for DESC

• Maps every project by matching its endpoint substations to OpenStreetMap, with a confidence grade for each location

• Finds 55 cross-utility overlaps within 40 km (25 mi), measured between the projects' closest points, and groups them into coordination tiers:

  Tier	What the utilities could share
  Touching / crossing	Must coordinate: outage timing, crossing structures
  Under 1.6 km	Right-of-way, access roads, permits
  Under 8 km	Laydown yards, deliveries
  Under 40 km	Crews and equipment

• Ranks the opportunities, with distance as the main signal and timeline overlap as the second

• Estimates the value of coordinating (shared mobilization and shared land) with editable assumptions, down to the savings per customer

• Explains each opportunity with Gemini: a short planner-ready brief built only from numbers the code computed

• Gets more accurate with use: anyone can confirm, correct or place a substation location. Verifications are stored in MongoDB Atlas and update everyone's map and rankings.

Key findings

#1 opportunity	DESC's new Jasper–Okatie 230 kV #2 line and Georgia Power's McIntosh–Purrysburg 230 kV project near Savannah: 2.98 mi apart, finishing 152 days apart. DESC's filing puts $23.8M of the project into its 2025 rate base.
Must-coordinate pair	DESC's Hooks–Thurmond tie and Georgia Power's Evans Primary–Thurmond Dam rebuilds end at the same substation.
Across all 55 pairs	About $1.2M in estimated coordination value (see assumptions)

Opportunity detail and cost estimate	Both utilities building into Thurmond Dam

How it works

 DATA PIPELINE (offline, results committed as CSV)
 ─────────────────────────────────────────────────
  DESC project PDF ─┐
                    ├─► ingest.py ─► geocode.py ─────────► projects.csv
  Georgia IRP PDF ──┘   (parse)      (OSM match, border
                                      checks, confidence)

 LIVE APP (Streamlit)
 ────────────────────
  projects.csv + MongoDB verifications ─► overlap.py ─► ranked opportunities
                                                              │
                   map + table  ◄─────────────────────────────┤
                   cost.py estimate ◄─────────────────────────┤
                   Gemini brief (explains computed facts) ◄───┘

Step	Module	What it does
Parse	gridlock/ingest.py	pdfplumber reads DESC's one-project-per-page list; pypdf reads the 10-year project table (pp. 177–190) of Georgia Power's 668-page IRP volume
Locate	gridlock/geocode.py	Extracts endpoint names ("SAV: GOSHEN (SAV) - MCINTOSH 115KV…" → Goshen, McIntosh), matches them to OSM substations (exact, then fuzzy), falls back to Nominatim, and rejects same-named substations in the wrong state using a Savannah River / NC / FL border model
Overlap	gridlock/overlap.py	Closest-point distance between project geometries in an equal-area projection (EPSG:5070), tiers, timeline gap, score
Cost	gridlock/cost.py	Shared mobilization and right-of-way savings, per-customer impact, rate-base context
Explain	gridlock/brief.py	Gemini brief grounded in computed facts, with fallback across models and cached results
Verify	gridlock/store.py	Community location verifications in MongoDB Atlas, layered on top of the automated matches
App	app.py	Streamlit + Folium: linked map, ranked table, detail panel, filters

How the numbers are calculated

Score (0–100): distance is the primary signal, timing the secondary one, discounted for uncertain locations:

$$\text{score} = w\left[45\left(1-\frac{d}{40\ \text{km}}\right) + \tfrac{1}{2}P + 35\max\left(0,\ 1-\frac{\Delta t}{1095\ \text{days}}\right)\right]$$

• d is the closest distance between the two projects.
• P is the tier bonus: 40, 30, 15 or 0 points.
• Δt is the gap between in-service dates.
• w is the location-confidence weight: 1.0 for high or verified, 0.9 for medium, 0.7 for low.

Savings:

• Shared mobilization = smaller project's cost × 8% × tier share (0.5 / 0.4 / 0.25 / 0.1) × timing factor.
  • The timing factor is 1.0 if the finish dates are within 6 months, 0.7 within 1 year, 0.4 within 2 years, and 0.1 beyond that.
• Shared right-of-way applies only to pairs within 1.6 km: parallel length × ROW width (30 / 38 / 61 m by voltage) × easement cost per acre.

Redacted costs: Georgia's public filing redacts all dollar amounts, so the app estimates them and labels them as estimates:

• $1.5M per mile for line projects, editable in the app.
• $10M for single-site projects such as substations.

Using the smaller project's cost means these estimates can't inflate the savings. For comparison, DESC's own filed lines cost about $2–2.2M per mile.

Validation

pytest runs 19 tests, including checks that reproduce all 6 overlaps in Sperry's answer key (center-to-center distances within 0.3 mi), plus the parser, the state-border model, tiers, and verification logic.

Location confidence: 197 of 252 projects are placed on the map.

Grade	Meaning
high	Exact substation match, and the operator matches the utility
medium	Exact name match, but no operator listed
low	Fuzzy or place-name match
verified	Confirmed or corrected by a user

The 55 projects without a location are mostly in the Atlanta area, far from the border.

The hackathon

Event	ShellHacks 2026, Florida's largest hackathon (10th anniversary), Florida International University, Sept 26–27, 2026
Team	Solo, first-time hacker
Challenge	Sperry Tech GridLock: compare at least two utilities' public construction plans, flag geographic (< 40 km) and timeline overlaps, show them on an interactive map with a ranked list, and estimate the impact (bonus)
Tracks entered	Sperry Tech GridLock · Microsoft "What's Missing?" · MLH Best Use of Gemini API · MLH Best Use of MongoDB Atlas · MLH Best Use of DigitalOcean · MLH Best Domain Name from GoDaddy Registry · Best First-Time Hacker
Deployed during the event	DigitalOcean App Platform (auto-deploy from this repo via Procfile), MongoDB Atlas, custom domain gridneighbors.design

Run it locally

python -m venv .venv
.venv/Scripts/pip install -r requirements.txt   # macOS/Linux: .venv/bin/pip
.venv/Scripts/streamlit run app.py

Optional features: copy .env.example to .env and fill in:

Variable	Enables
GEMINI_API_KEY	Gemini coordination briefs (Google AI Studio)
GEMINI_MODEL	Preferred Gemini model (default gemini-3.8-flash)
MONGODB_URI	Shared location verifications (MongoDB Atlas connection string)

The app runs without either one; those features are simply hidden.

Project structure

app.py                  Streamlit app (map, ranked table, detail panel, verification, briefs)
gridlock/
  ingest.py             Parse the two utilities' PDF project lists
  geocode.py            Match endpoint substations to OpenStreetMap; border model; confidence
  overlap.py            Closest-point overlaps, tiers, timeline gap, score
  cost.py               Savings and per-customer impact estimate
  brief.py              Gemini coordination briefs (with model fallback and cache)
  store.py              MongoDB Atlas location verifications
  pregen_briefs.py      Pre-generate briefs for the top opportunities
data/
  processed/            projects_raw.csv, projects.csv, overlaps.csv
  overrides.csv         Hand-verified substation coordinates
  briefs.json           Cached Gemini briefs
tests/                  pytest suite (includes Sperry's answer key)
tools/capture_media.py  Headless-browser capture of the screenshots and demo video
docs/images/            README images

Rebuilding the data

The processed CSVs are committed, so this is only needed to regenerate them. It requires the sponsor's Sperry-Tech-Challenge/ folder, which is not committed, and an OpenStreetMap substation export saved to data/cache/osm_substations.json. That export comes from this Overpass query:

[out:json][timeout:180];
( nwr["power"="substation"](30.3,-85.7,35.3,-78.4);
  nwr["power"="plant"](30.3,-85.7,35.3,-78.4); );
out center tags;

.venv/Scripts/python -m gridlock.ingest     # PDFs -> data/processed/projects_raw.csv
.venv/Scripts/python -m gridlock.geocode    # -> data/processed/projects.csv (Nominatim fallback is cached)
.venv/Scripts/python -m gridlock.overlap    # -> data/processed/overlaps.csv
.venv/Scripts/python -m pytest

To hand-verify coordinates, add them to data/overrides.csv. They're zone-aware, since Georgia has two "Goshen" substations.

Limitations and what's next

• Straight lines: utilities publish endpoint substations, not routes, so lines are drawn straight between them. Next: use real OSM transmission-line routes.
• Open editing: anyone can submit a location verification (every edit is logged with the previous location). Next: sign-in for utility and regulator staff, and a review step.
• Two utilities: the pipeline doesn't depend on a particular utility, so the next step is adding more neighbors (Duke, Santee Cooper).
• Customer view: a ZIP-code view showing the grid projects near you and how they reach your electric bill.

Built with

Python · pandas · pdfplumber · pypdf · Shapely · pyproj · OpenStreetMap (Overpass, Nominatim) · Streamlit · Folium · Google Gemini API · MongoDB Atlas · pytest · Playwright. Hosted during the event on DigitalOcean App Platform with a GoDaddy Registry (.design) domain.

Acknowledgments

Built by Jonathan Guerrero at ShellHacks 2026 (FIU), my first hackathon. Thanks to Sperry Tech for the GridLock challenge and its data package, and to the ShellHacks organizers and MLH. Developed with AI coding assistance (Claude Code).

Data: Dominion Energy South Carolina Planned Transmission Projects $2M and above (SCRTP) and the Georgia ITS 10-Year Plan in Georgia Power's 2025 IRP Vol. 3, public disclosure version. Only public filings are used. Map data © OpenStreetMap contributors.