EVAT - Electric Vehicle Adoption Tools

A conversational AI chatbot designed to help electric vehicle users find charging stations and get relevant information using Rasa framework.

🚀 Features

• Location-based charging station finder
• Real-time traffic-aware routing
• Charging station filtering by preferences
• Detailed station information
• Web-based chat interface
• Planned: live connector availability, ML-powered ETA

📁 Project Structure

EVAT_Chatbot/
├── rasa/                 # Rasa chatbot configuration
│   ├── domain.yml       # Intent, entity, and action definitions
│   ├── config.yml       # NLU pipeline and policy configuration
│   ├── endpoints.yml    # API endpoints
│   ├── credentials.yml  # Authentication settings
│   ├── actions/         # Custom action implementations
│   └── data/           # Training data (intents, stories, rules)
├── backend/            # Core business logic
│   ├── real_time_apis.py # TomTom client used by actions
│   └── utils/          # Utility functions
├── frontend/           # Web interface
│   ├── index.html      # Main chat interface
│   ├── script.js       # Frontend logic
│   └── style.css       # Styling
├── data/              # Datasets
│   └── raw/           # CSV files (charging stations, coordinates)
├── ml/                # Machine learning models
│   ├── classification.py # Station classification
│   ├── regression.py    # ETA prediction
│   └── README.md       # ML documentation
├── config/            # Configuration files
├── README.md          # Project overview
├── requirements.txt   # Dependencies
└── .gitignore        # Git ignore rules

🧩 How to use the chatbot (local setup)

Quick setup

cd EVAT_Chatbot
python -m venv rasa_env && source rasa_env/bin/activate
pip install -r requirements.txt
cd rasa && rasa train

Tab 1: rasa run actions --port 5055
Tab 2: rasa run --enable-api --cors "*"
Open frontend/index.html (or serve frontend/ via python -m http.server 8080)

Open the web chat

• Open frontend/index.html in your browser. It posts to http://localhost:5005/webhooks/rest/webhook.
• Ensure the Rasa server was started with --cors "*" so the browser can call it.

cd frontend
python -m http.server 8080
# open http://localhost:8080

Interact (3 flows)

• Route planning: 1 → from Carlton to Geelong → get_directions pls → get_traffic_info
• Emergency charging: 2 → Richmond → type a station name → get directions pls
• Preferences: 3 → fastest (or cheapest/premium) → Melbourne → type a station name

How it works

• Locations resolve from CSV names to coordinates; if a name isn’t in the CSV, you’ll be asked to try another.
• TomTom provides real-time distance/ETA/traffic when both start and destination coords exist.
• The frontend is wired to Rasa REST and also sends browser geolocation (lat, lon) as metadata; actions currently do not use this metadata yet.

Real-time TomTom

Handled via backend/real_time_apis.py (used by Rasa actions)

🖥️ Frontend (current state)

• The chat UI in frontend/ is already wired to the Rasa REST webhook.

• It sends lat/lon from the browser as metadata with each message.

• Limitations today:

  • Actions do not yet use the metadata.lat/lon to improve results.
  • Plain text rendering only (no quick-reply buttons or cards yet).
  • Live availability is not implemented.

📍 Data sources used

• data/raw/Co-ordinates.csv for suburb coordinates
• data/raw/charger_info_mel.csv for stations

🔄 Real-time readiness

• Done:
  • Key in .env(TOMTOM_API_KEY) used by backend
  • Real-time routing + traffic via TomTom (CSV-backed locations)
  • Frontend wired via REST; or use Rasa shell
• How it works now:
  • Names (start/end) resolve to coordinates via CSV only; then TomTom provides route/traffic
  • Frontend sends metadata lat/lon, but actions currently resolve start/end from CSV
• Gaps:
  • Start from user location (metadata lat/lon) not yet used by actions
  • Stations are CSV-based (no TomTom station search or availability yet)
  • Dataset: missing/null values; some station names/addresses inconsistent
  • Matching: station lookup is strict; fuzzy matching can be added for better tolerance
• Next:

1. Use browser lat/lon (metadata) as start coords; keep CSV for names.
2. Add along-route/nearby station search (TomTom) as fallback when CSV returns no results.
3. Add live availability (new API) and fold into ranking.
4. Add fuzzy matching + CSV cleanup.
5. Frontend: UI enhancements and interactivity — quick-reply buttons (send payloads), clickable options, station “cards” with details and CTAs (Get directions, Show traffic)...
6. Optional: incorporate ML ranking/ETA once wired into actions.