AI storm operations copilot for utilities and microgrids.
GridShadow turns live weather into a ranked risk view and generates a personalized storm preparation + restoration plan grounded in operator runbooks, constraints, and priorities.

During extreme weather, operators receive forecasts and alerts, but translating that into decisions is still manual. GridShadow helps answer:

• What zones/feeders/assets are most likely to fail in the next 6–48 hours?
• Where should crews and resources be staged before impact?
• What should be restored first during outages?
• How do we prioritize critical infrastructure (hospitals, shelters, EOCs, water) without delaying overall restoration?

GridShadow is built to be operator-grade: it produces checklists, dispatch tasks, and exportable briefs, not just charts.

• Live storm awareness: Forecast + alerts timeline for a selected territory
• Risk ranking: Top at-risk zones/feeders/assets with driver explanations
• Critical-first planning: Adjustable tradeoff between:
  • maximize customers restored and
  • critical facilities first
• Supermemory personalization: Ingests operator documents (runbooks, priority rules, crew resources, incident logs) so the same storm produces different plans for different operators
• Exportable briefs: Control room ops brief, crew dispatch task list, public safety brief
• Learning loop: After-action logging updates vulnerability notes and procedural learnings

1. Situation: live forecast + active alerts + storm timeline
2. Risk: ranked high-risk zones + critical facilities at risk
3. Plan: storm prep checklist + crew staging + restoration order
4. What-if: slider to shift objective between critical-first vs max-customers
5. Copilot: structured responses with actions, reasons, triggers, assumptions
6. After-action: log outcomes → memory updates → next run improves

Forecast → Risk → Plan → Brief → Learn

We use lightweight reasoning components to avoid generic chatbot output:

• Retriever: pulls the most relevant runbook sections / thresholds / priorities from memory
• Risk scorer: converts storm signals into risk scores with explainable drivers
• Planner: generates prep + restoration steps under constraints (crews, SOC floors, islanding rules)
• Verifier: checks for missing steps, approvals, and safety constraints; outputs assumptions + triggers

• Frontend: Next.js 16 (React 19)
• Backend: Next.js API routes (Node.js)
• Database: PostgreSQL (Cloud SQL)
• LLM: Gemini API
• Memory: Supermemory (RAG)
• Weather: Google Weather API
• Maps / Geocoding: Google Maps Platform
• Visualization: Recharts, D3, Google Maps

git clone https://github.com/kbhatnagar1506/GRIDSHADOW.git
cd GRIDSHADOW

pnpm install

cp .env.example .env.local

Edit .env.local and set at least:

• DATABASE_URL — PostgreSQL (see DEPLOY.md for Cloud SQL)
• SUPERMEMORY_API_KEY — for connectors and Copilot memory
• GEMINI_API_KEY or GOOGLE_AI_API_KEY — for Copilot (optional; uses demo responses if unset)
• GOOGLE_MAPS_API_KEY or GOOGLE_WEATHER_API_KEY — for weather
• NEXT_PUBLIC_GOOGLE_MAPS_API_KEY — for the Situation map

Run scripts/schema-multitenant.sql (or scripts/migrate-tenant-profiles.sql) on your Postgres instance.

pnpm dev

Open http://localhost:3000.

• Heroku: See DEPLOY.md for Cloud SQL, config vars, and git push heroku main.
• Live: pacific-wildwood-03683-e30507b8ccff.herokuapp.com