CascadeIQ

CascadeIQ is a 3D grid platform that simulates cascade failures in Atlanta, scores real-time risk with ML, and uses AI to capture critical maintenance knowledge from retiring engineers.

Comment

CascadeIQ

3D Infrastructure Resilience Simulation Platform
Cox Play With Purpose Hackathon — Energy Track, Southern Company, Atlanta GA

What It Does

CascadeIQ is a real-time grid intelligence platform that simulates cascade failures across Atlanta's and Macon's energy infrastructure. It combines live risk scoring, AI-powered analysis, and an institutional knowledge layer to answer the question: when a transformer fails, what goes dark — and does anyone know how to fix it?

Core capabilities:

  • Interactive 3D map of 2,535 real energy assets (transformers, MARTA stations, hospitals, solar farms, data centers, switching stations, and more)
  • Cascade failure simulation — click any transformer to see which hospitals, MARTA stations, and traffic signals go dark
  • ML risk scoring (XGBoost + rule-based hybrid) updated in real time
  • AI agent (Claude) for natural language infrastructure queries
  • Knowledge Intelligence Layer — 20+ years of simulated maintenance records, incident reports, and engineer retirement risk analysis
  • Build mode — simulate deploying solar farms, microgrids, and BESS to reduce network risk

Quick Start

Prerequisites: Python 3.10+, Node 18+, Anthropic API key, Mapbox token

1. Clone and set up environment

git clone <repo>
cd InfraIQ

Create C:\InfraIQ\.env:

ANTHROPIC_API_KEY=sk-ant-...

Create C:\InfraIQ\frontend\.env:

VITE_MAPBOX_TOKEN=pk.eyJ1...
VITE_BACKEND_URL=http://localhost:8080

2. Install backend dependencies

cd backend
pip install -r requirements.txt

3. Start the backend

cd C:\InfraIQ\backend
python -m uvicorn main:app --host 0.0.0.0 --port 8080

When you see Application startup complete, the server is ready.
API docs: http://localhost:8080/docs

4. Start the frontend

cd C:\InfraIQ\frontend
npm install
npm run dev

App: http://localhost:5173 (or next available port)

Architecture

InfraIQ/
├── backend/
│   ├── main.py              # FastAPI app, all 14+ endpoints, port 8080
│   └── knowledge_engine.py  # Knowledge extraction engine (Claude + rule-based)
├── frontend/
│   └── src/
│       ├── App.jsx               # Root — state, asset selection, demo flow
│       ├── Map.jsx               # Mapbox GL 3D map layer
│       ├── SidePanel.jsx         # Right panel — risk scores, cascade impact
│       ├── KnowledgePanel.jsx    # Left panel — Timeline / Knowledge / Ingest tabs
│       ├── KnowledgeDashboard.jsx # Full overlay — 4-section knowledge hub
│       ├── BuildMode.jsx         # Infrastructure build simulation UI
│       ├── AgentInput.jsx        # Natural language AI query bar
│       └── api.js                # All backend API calls
├── data/
│   ├── raw/
│   │   ├── maintenance_records.json  # 300 work orders (2003–2026)
│   │   ├── incident_reports.json     # 40 incidents (2020–2024)
│   │   ├── retirement_records.json   # 8 engineers
│   │   └── procedures.json           # 25 SOPs (11 outdated)
│   └── processed/
│       ├── assets.json          # 2,535 grid assets
│       ├── assets.geojson       # GeoJSON for map rendering
│       └── graph.json           # Dependency graph for cascade simulation
└── scripts/
    ├── fetch_new_assets.py      # Pulls from Overpass API (OSM)
    └── gen_company_data_final.py # Generates all simulated records

API Endpoints

All endpoints at http://localhost:8080

Method Path Description
GET /health Server status + asset count
GET /assets GeoJSON of all 2,535 assets + map_config
GET /stats Network health grade, cost exposure, by-type breakdown
GET /top-risks Top 10 highest-risk transformers
GET /cascade/{asset_id} Cascade failure simulation for an asset
POST /optimize Budget optimization for infrastructure investments
POST /agent AI agent — natural language queries
GET /asset/{asset_id}/knowledge Knowledge for a specific asset
GET /knowledge/stats Overall knowledge health metrics
GET /knowledge/blindspots Assets with zero documentation
GET /knowledge/retirement-risk Retiring engineers and at-risk assets
GET /knowledge/patterns Cross-asset patterns discovered by AI
GET /knowledge/procedures SOPs with outdated procedure detection
GET /knowledge/timeline/{asset_id} Chronological event history for an asset
POST /knowledge/ingest Paste any document; AI extracts and attaches knowledge

Knowledge Intelligence Layer

The Knowledge Intelligence Layer addresses a real problem in utility companies: critical operational knowledge lives in engineers' heads, not in any system. When a 38-year veteran retires, so does the knowledge of which transformer shares a grounding bus with MARTA, which switching sequence takes 11 hours instead of 3, and which units can safely run at 115% nameplate for 4 hours.

What it does

  • 300 maintenance records spanning 2003–2026, authored by 10 technicians with realistic field notes
  • 40 incident reports including the 2021 Atlanta heat dome, 2022 Hurricane Ian, 2023 ice storm, and 2024 data center overload
  • 8 retirement records tracking knowledge risk as engineers leave (Dave Morrison: 8% captured; James Harrison: 5% captured)
  • 25 SOPs, 11 flagged as outdated (referencing discontinued equipment: Westinghouse BG-3, GE HV-1, ABB type 27)
  • Knowledge extraction engine (knowledge_engine.py) matches records to assets by lat/lon proximity (500m radius), extracts patterns using Claude, identifies blind spots, and scores knowledge risk

How to use in the UI

  1. Click any asset on the map — the Knowledge Panel slides in from the left (Timeline tab shows chronological history; Knowledge tab shows AI-extracted patterns and health trajectory; Ingest tab lets you paste any document for AI extraction)
  2. Click "🧠 Knowledge Hub" in the header — opens the full dashboard showing coverage grade, retiring engineers, most dangerous blind spots, and cross-asset patterns
  3. Red assets in Knowledge View = blind spots (zero documentation + high failure risk)

Demo Flow

The panel shortcut Ctrl+Shift+D (or Cmd+Shift+D) runs an automated demo sequence.

Manual demo for judges:

  1. Map loads with 2,535 assets across Atlanta and Macon
  2. Click "🧠 Knowledge Hub" → show the F-grade knowledge coverage (only 14.3% of transformers documented)
  3. See "Dave Morrison retiring" — 38 years experience, 8% captured, 47 assets at risk
  4. Close dashboard, click a red transformer on the map
  5. Right panel: high failure probability, cascade impacts 7 MARTA stations
  6. Left panel (Timeline tab): scroll through 20 years of field notes — senior engineers flagging the same thermal issue every August since 2008
  7. Left panel (Knowledge tab): AI-extracted pattern — "recurring thermal stress, unit undersized for 1987 load profile"
  8. Left panel (Ingest tab): paste a work order → AI extracts facts and attaches to asset
  9. Ask the agent: "Which transformers need urgent action before summer?"

Data Sources

  • Asset locations: OpenStreetMap via Overpass API (transformers, substations, solar farms, MARTA stations, hospitals, power plants, data centers, EV charging)
  • Risk scores: XGBoost ML model + rule-based scoring (age, load, thermal history, cascade score)
  • Maintenance records: Simulated from realistic Southern Company operational patterns
  • Geographic coverage: Atlanta Metro + Macon GA (statewide Georgia view auto-detected)

Team

Built for the Cox Play With Purpose Hackathon — Energy Track, powered by Southern Company infrastructure data and the Atlanta grid.

Share this project:

Updates

Submission history