GridWise ⚡

AI-Powered Infrastructure Intelligence Platform

Detect defects. Generate reports. Predict failures. Before they happen.

Built as a deep-dive MVP inspired by ENRZY — the infrastructure intelligence platform by BetterDrones India. GridWise mirrors their 5-step pipeline and adds failure risk forecasting — predicting asset failures before they happen.

Features • Architecture • API Reference • Quick Start • Demo

---

🌍 The Real-World Problem

India's power grid has 4.5 lakh+ transmission towers spanning hundreds of thousands of kilometres. Every tower needs regular inspection. Traditionally, this meant:

• 👷 A human inspector physically climbing or visiting each tower
• 📋 Handwritten notes, manual reports, emailed PDFs
• 🐌 Days between inspection and maintenance action
• 🔴 Zero prediction — defects found only after they've developed

BetterDrones + ENRZY changed this — drones now fly the lines, AI detects the defects, and reports are generated in minutes.

GridWise takes it further — by analysing historical inspection data to predict which towers will fail in the next 30, 60, or 90 days. From reactive to proactive. From damage control to decision intelligence.

---

✨ Features

🛡️ DataGuard — Smart Upload Validation

No bad data reaches the AI. Ever.

• Validates image resolution, format, and completeness before analysis
• Rejects substandard files and notifies the pilot to re-shoot
• Attaches geo-tags from EXIF or request metadata
• Every inspection gets a clean, verified data package

🤖 AI Defect Detection

Claude vision analyses every image for infrastructure defects.

Detects across 7 defect categories:

Defect	Severity Range
🟥 Corrosion / rust	Critical → Minor
🌿 Vegetation encroachment	Critical → Minor
🔩 Missing components	Critical → Major
↘️ Sag / clearance violation	Critical → Major
🔓 Cracks / fractures	Critical → Minor
🌡️ Thermal hotspots	Critical → Major
⚠️ Other anomalies	Major → Minor

Returns structured JSON with defect type, severity, exact location on asset, confidence score, and AI reasoning — all stored against the inspection.

📄 Automated Report Generation

From raw defects to a client-ready PDF in seconds.

• LLM writes executive summary, health score justification, and recommendations
• Health score (0–100) calculated per inspection
• PDF rendered with asset info, defect table, severity colour coding, and annotated findings
• Ready to send directly to PGCIL or any utility client

🔧 FixTrack — Maintenance Ticket Engine

Every defect becomes a work order. Automatically.

• One ticket auto-created per defect
• Priority set by severity: P1 (Critical, 7 days) → P2 (Major, 21 days) → P3 (Minor, 45 days)
• AI writes field-ready step-by-step instructions per ticket
• Before/after repair photo upload to verify and close tickets
• Status tracking: open → in_progress → closed

🔮 Failure Risk Forecasting (Original Feature)

ENRZY detects defects. GridWise predicts failures.

This is what ENRZY doesn't have yet. GridWise feeds an asset's full inspection history into Claude and asks it to reason about the degradation trend:

• Failure probability at 30 / 60 / 90 days
• Degradation rate in health score points per month
• Which specific component is most at risk
• Recommended action: monitor / schedule maintenance / emergency intervention
• Full reasoning chain — explainable AI, not a black box

A tower going from health score 88 → 74 → 61 → 52 → 41 over 12 months doesn't just need a ticket — it needs a team on-site before next quarter.

---

🏗️ Architecture

┌─────────────────────────────────────────────────────────┐
│                      Client Layer                       │
│          React UI  /  Postman  /  curl / SDK            │
└────────────────────────┬────────────────────────────────┘
                         │ HTTP
┌────────────────────────▼────────────────────────────────┐
│                   FastAPI Backend                       │
│                                                         │
│  ┌─────────────┐  ┌─────────────┐  ┌────────────────┐  │
│  │  DataGuard  │  │ AI Detection│  │ Report Engine  │  │
│  │  Validate   │  │Claude Vision│  │  LLM → PDF     │  │
│  │  Geo-tag    │  │Defect → JSON│  │  Health Score  │  │
│  └─────────────┘  └─────────────┘  └────────────────┘  │
│                                                         │
│  ┌─────────────────────────────────────────────────┐    │
│  │         FixTrack Ticket Engine                  │    │
│  │  Defect → Ticket → Assign → Verify → Close     │    │
│  └─────────────────────────────────────────────────┘    │
│                                                         │
│  ┌─────────────────────────────────────────────────┐    │
│  │      Failure Risk Forecaster  🔮                │    │
│  │  Inspection history → Claude reasoning → Risk   │    │
│  └─────────────────────────────────────────────────┘    │
└────────────────────────┬────────────────────────────────┘
                         │
┌────────────────────────▼────────────────────────────────┐
│                  Data + AI Layer                        │
│                                                         │
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐  │
│  │  PostgreSQL  │  │ File Storage │  │  Claude API  │  │
│  │  4 tables    │  │ Images/PDFs  │  │ Vision+Text  │  │
│  │  + forecasts │  │ local / S3   │  │   Sonnet 4.5 │  │
│  └──────────────┘  └──────────────┘  └──────────────┘  │
└─────────────────────────────────────────────────────────┘

📁 Project Structure

gridwise/
├── app/
│   ├── main.py                 # FastAPI app entry point
│   ├── config.py               # Environment config
│   ├── database.py             # Async SQLAlchemy setup
│   ├── models/                 # ORM models
│   │   ├── asset.py
│   │   ├── inspection.py
│   │   ├── defect.py
│   │   ├── ticket.py
│   │   └── forecast.py
│   ├── schemas/                # Pydantic v2 schemas
│   ├── routers/                # FastAPI route handlers
│   │   ├── assets.py
│   │   ├── inspections.py
│   │   ├── tickets.py
│   │   └── forecasts.py
│   ├── services/               # Business logic
│   │   ├── dataguard.py
│   │   ├── ai_detection.py
│   │   ├── report_engine.py
│   │   ├── ticket_engine.py
│   │   └── forecaster.py
│   ├── ai/
│   │   ├── client.py           # Anthropic client
│   │   ├── prompts.py          # All prompt templates
│   │   └── parsers.py          # Safe JSON parsers
│   └── storage/
│       └── file_handler.py     # File I/O (S3-ready)
├── alembic/                    # DB migrations
├── tests/                      # Pytest test suite
├── demo_seed.py                # One-command demo setup
├── .env.example
├── requirements.txt
└── README.md

---

🗄️ Database Schema

assets                          inspections
──────────────────────          ──────────────────────────────
id          UUID PK             id               UUID PK
name        VARCHAR             asset_id         UUID FK → assets
asset_type  ENUM                pilot_id         VARCHAR
latitude    FLOAT               capture_date     DATE
longitude   FLOAT               capture_types    ARRAY
zone        VARCHAR             validation_status ENUM
installed_year INT              health_score     INT
created_at  TIMESTAMP           created_at       TIMESTAMP


defects                         tickets
──────────────────────────      ──────────────────────────────
id             UUID PK          id               UUID PK
inspection_id  UUID FK          defect_id        UUID FK
defect_type    ENUM             asset_id         UUID FK
severity       ENUM             priority         ENUM  P1/P2/P3
location_desc  TEXT             status           ENUM
confidence     FLOAT            title            VARCHAR
ai_reasoning   TEXT             instructions     TEXT
created_at     TIMESTAMP        assigned_team    VARCHAR
                                due_date         DATE
                                before_photo     VARCHAR
forecasts                       after_photo      VARCHAR
──────────────────────────      closed_at        TIMESTAMP
id             UUID PK
asset_id       UUID FK
risk_30_days   FLOAT
risk_60_days   FLOAT
risk_90_days   FLOAT
degradation_rate FLOAT
at_risk_component VARCHAR
recommended_action VARCHAR
reasoning      TEXT
generated_at   TIMESTAMP

---

🚀 Quick Start

Prerequisites

• Python 3.11+
• PostgreSQL 16+
• An Anthropic API key

1. Clone and install

git clone https://github.com/0xvicky/gridwise.git
cd gridwise
python -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate
pip install -r requirements.txt

2. Configure environment

cp .env.example .env

Edit .env:

DATABASE_URL=postgresql+asyncpg://postgres:password@localhost:5432/gridwise
ANTHROPIC_API_KEY=sk-ant-...
CLAUDE_MODEL=claude-sonnet-4-5
STORAGE_PATH=./storage
MAX_IMAGE_SIZE_MB=50
MIN_IMAGE_WIDTH=4000
MIN_IMAGE_HEIGHT=3000

3. Set up the database

alembic upgrade head

4. Seed demo data

python demo_seed.py

This creates 3 assets with 5 inspections each showing a realistic degradation curve — ready for the forecaster to analyse.

5. Run the server

uvicorn app.main:app --reload

API docs live at http://localhost:8000/docs ✨

---

📡 API Reference

Assets

Method	Endpoint	Description
POST	/assets	Register a new infrastructure asset
GET	/assets	List all assets (filter by zone, health score)
GET	/assets/{asset_id}	Asset detail + full inspection history

DataGuard

Method	Endpoint	Description
POST	/inspections/upload	Upload drone images + metadata
GET	/inspections/{id}/validation	Get per-file validation results
POST	/inspections/{id}/reupload	Re-submit rejected images

AI Detection

Method	Endpoint	Description
POST	/inspections/{id}/analyze	🤖 Run Claude vision defect detection
GET	/inspections/{id}/defects	Get all detected defects

Reports

Method	Endpoint	Description
POST	/inspections/{id}/report	📄 Generate PDF inspection report
GET	/inspections/{id}/report	Download the generated PDF

FixTrack

Method	Endpoint	Description
POST	/inspections/{id}/tickets/generate	🔧 Auto-create maintenance tickets
GET	/tickets	List all tickets (filter by status/priority)
GET	/tickets/{ticket_id}	Get single ticket detail
PATCH	/tickets/{ticket_id}/status	Update ticket status
POST	/tickets/{ticket_id}/repair-photo	Upload before/after repair photo

Forecaster

Method	Endpoint	Description
POST	/assets/{id}/forecast	🔮 Run failure risk forecast
GET	/assets/{id}/forecast	Get latest forecast
GET	/forecasts/high-risk	Assets with >70% failure risk in 30 days

---

🎬 The 5-Minute Demo

After running demo_seed.py, walk through this story:

# 1. Check Tower T-482 — health declining from 88 → 41 over 5 inspections
GET /assets/tower-t482

# 2. Upload a new inspection image
POST /inspections/upload
  → inspection_id: abc-123

# 3. Validate passed — all images clean
GET /inspections/abc-123/validation
  → status: passed, 12/12 images accepted

# 4. Run AI detection
POST /inspections/abc-123/analyze
  → defect: corrosion, severity: critical, confidence: 0.87
  → defect: vegetation_encroachment, severity: major, confidence: 0.91

# 5. Generate PDF report
POST /inspections/abc-123/report
  → health_score: 38/100
  → PDF downloaded with full defect analysis

# 6. Auto-create maintenance tickets
POST /inspections/abc-123/tickets/generate
  → TKT-001: P1 Corrosion repair — due in 7 days
  → TKT-002: P2 Vegetation clearance — due in 21 days

# 7. Close ticket with repair photo
POST /tickets/TKT-001/repair-photo   (upload after.jpg)
PATCH /tickets/TKT-001/status        { "status": "closed" }

# 8. Run failure risk forecast
POST /assets/tower-t482/forecast
  → risk_30_days: 0.84
  → recommended_action: "Emergency intervention required"
  → reasoning: "Asset degrading at 9.4 points/month..."

# 9. See the watchlist
GET /forecasts/high-risk
  → Tower T-482 at top — 84% failure risk

---

🧪 Running Tests

pytest tests/ -v

tests/test_assets.py::test_create_asset           PASSED ✅
tests/test_inspections.py::test_upload_and_validate PASSED ✅
tests/test_inspections.py::test_dataguard_rejects_blurry PASSED ✅
tests/test_ai_detection.py::test_defect_detection  PASSED ✅
tests/test_tickets.py::test_auto_ticket_generation PASSED ✅
tests/test_tickets.py::test_cannot_close_without_photo PASSED ✅
tests/test_forecaster.py::test_forecast_output     PASSED ✅
tests/test_forecaster.py::test_forecast_needs_min_2_inspections PASSED ✅

---

🔒 Business Rules

These are enforced at the API layer — not just convention:

• 🚫 DataGuard gate — /analyze returns HTTP 400 if inspection is not validated
• 📸 Repair photo required — tickets cannot close without an after-photo uploaded
• 📊 Minimum history — forecaster requires at least 2 inspections with health scores
• 🔑 No hardcoded keys — all secrets via .env only
• 📁 Relative file paths — storage is S3-ready, no absolute paths in DB

---

🛣️ What's Next

• WebSocket real-time ticket status updates
• Multi-tenant support (separate orgs per utility client)
• LiDAR point cloud ingestion + 3D model generation
• Mobile app for field crews (repair photo upload in the field)
• Scheduled re-forecast jobs — auto-run forecaster weekly per asset
• Slack / WhatsApp alerts when a tower crosses 70% failure risk threshold

---

🤝 Built For

This MVP was built to demonstrate deep understanding of the infrastructure intelligence problem space being solved by

BetterDrones and their platform ENRZY

India's leading drone-powered infrastructure intelligence company.

---

📄 License

MIT © 2025 GridWise

---

Built with ⚡ by Vivek Tyagi

"Quality Data — On Time — Every Time"