Building an EV Charging Station Predictor & AI Assistant at Exicom

During my internship at Exicom, I worked on developing two key components to enhance the electric vehicle (EV) charging infrastructure:

1. An intelligent predictor system for optimal EV charging station locations
2. An AI-powered chatbot assistant for EV-related queries

EV Charging Station Location Predictor

The location predictor system helps identify optimal locations for new EV charging stations by analyzing various factors:

• Proximity to key amenities (restaurants, cafes, malls, etc.)
• Parking availability
• Public transport accessibility
• Highway service areas
• Population density

Technical Implementation

• Built using Python with Flask backend
• Utilized machine learning (RandomForestRegressor) for location scoring
• Integrated OpenStreetMap data via OSMnx for amenity analysis
• Implemented geospatial analysis using GeoPandas
• Created interactive visualizations with Folium maps

Key Features

• Suitability score prediction for specific locations
• Region-wide analysis with multiple point sampling
• Customizable feature selection
• Interactive map visualizations
• Detailed proximity analysis for amenities
• CSV export functionality for analysis results

The Magic Behind Location Prediction

Let me share how our prediction system actually works - it's pretty fascinating! Think of it like a smart real estate agent who knows exactly what makes a location perfect for an EV charging station.

Smart Feature Analysis

The system looks at ten key factors that make a location ideal:

• Restaurants and cafes (because who doesn't grab a coffee while charging?)
• Shopping areas like malls and supermarkets
• Parking facilities (obviously crucial!)
• Public transport hubs
• Hotels and tourist spots
• Highway service areas

Each of these factors gets a different weight in our calculations. For example, parking availability and mall proximity are given higher importance (15% each) compared to bicycle parking (5%) - because let's face it, most EV owners are more likely to need a parking spot than a bike rack!

The Distance Magic

Here's where it gets interesting - we don't just check if these amenities exist, but how far they are. The system uses something called "exponential decay" (fancy term, I know!) which basically means that a restaurant 100 meters away is much more valuable than one 1 kilometer away. It's like real estate's "location, location, location" mantra, but with mathematical precision!

Making Sure It's Actually Usable

We also do some practical checks:

• Is the location actually on land? (Yes, we had to add this check after the system once suggested a charging station in the middle of a lake! 😅)
• Are there enough potential customers nearby?
• Is it accessible from major roads?

The Learning Part

The really cool part is how the system learns. We fed it data about successful charging stations and their surroundings, and it learned the patterns. Using a machine learning model called Random Forest (think of it as a super-smart decision tree that can handle lots of factors at once), it can now look at any location and say "This spot is 85% suitable for a charging station!"

Visual Insights

My favorite part is the visualization system. For any location you pick, it creates an interactive map showing:

• Green circles for nearby amenities (the good stuff)
• Yellow for things that are a bit far
• Red for amenities that are too distant

It's like having a heat map of convenience! You can literally see why a location is good or bad.

EVAI - The EV Assistant Chatbot

Alongside the predictor, I developed an AI-powered chatbot specialized in EV information:

Technical Stack

• Built using LangChain framework
• Integrated with Ollama for local model deployment
• Implemented structured logging and feedback collection
• Used pandas for data processing
• Added batch processing capabilities for efficient operation

Chatbot Capabilities

• Provides detailed information about EV charging stations
• Answers queries about charging compatibility
• Offers guidance on nearby amenities
• Maintains conversation context for better responses
• Collects and learns from user feedback

Learning Outcomes

This internship provided valuable experience in:

• Machine learning model development and deployment
• Large language model integration
• Geospatial data analysis
• API development with Flask
• Real-world software engineering practices
• Working with cross-functional teams

Impact

The project contributed to Exicom's mission of expanding EV infrastructure by:

• Streamlining the location selection process for new charging stations
• Improving user experience through AI assistance
• Providing data-driven insights for infrastructure planning
• Reducing the time and effort required for site analysis

Conclusion

This internship project combined multiple cutting-edge technologies to create practical solutions for the EV industry. It was an excellent opportunity to work with real-world data and create tools that contribute to sustainable transportation infrastructure.