ChargeSmart

One app to find the cheapest, nearest EV charger in China. Powered by AI and crowdsourced data.

Comment

Inspiration

As a Chinese international student in Australia, I noticed how easy it is to compare fuel prices here using apps like Petrol Spy. In contrast, when my family charges their EV in China, finding the best option often means switching between multiple operator apps like State Grid, TGOOD, and Star Charge, with no simple way to compare prices or station quality side by side.

That contrast inspired me to build ChargeSmart — a tool that makes public EV charging more transparent, easier to compare, and more actionable.

What it does

ChargeSmart is an AI-powered EV charging decision platform for mainland China.

It helps users:

  • discover nearby charging stations across operators on a single map,
  • compare fragmented charging options in one place,
  • get multi-factor recommendations based on price, distance, availability, reliability, and speed,
  • contribute trusted community updates such as prices, faults, and queue conditions,
  • scan charging receipts with AI to extract bill data and calculate carbon savings,
  • ask natural-language questions like “Where is the cheapest fast charger near me right now?”

Rather than acting as a simple station finder, ChargeSmart helps users answer a more useful question: Where should I charge right now, and why?

How we built it

  • Frontend: Expo React Native (SDK 54) with Mapbox GL JS embedded via WebView for live station visualization
  • Backend: Node.js + Fastify + TypeScript, deployed on Railway
  • Database: PostgreSQL with Drizzle ORM, seeded with 50 real Shenzhen charging station locations and reference pricing
  • AI: Featherless API using Qwen2.5-7B for natural-language charging assistance and Qwen3-VL-30B for vision-based receipt scanning
  • Auth: Phone OTP login system
  • Trust & anti-fraud: Geofencing (users must be within 300m of a station to report), cross-validation (2+ confirmations before updates go live), receipt date validation, daily limits, and points caps
  • Development process: Built as a solo project with Claude AI as a development assistant for ideation, implementation support, and debugging

Challenges we ran into

  • Fragmented operator data: Charging information exists, but it is spread across disconnected operator ecosystems and not easily accessible through unified public integrations. To validate the MVP, we seeded the system with real Shenzhen station locations and reference pricing.
  • Receipt fraud prevention: Early versions mistakenly accepted unrelated receipts. We improved this through a multi-layer validation pipeline requiring explicit charging-related fields such as kWh units, operator matching, and backend numeric checks.
  • Map integration: Because Mapbox GL JS does not run natively inside React Native, we embedded it through WebView and built a custom message bridge for station interactions and user location updates.

Accomplishments that we're proud of

  • Built and deployed a functioning full-stack MVP as a solo developer in under two weeks
  • Created a multi-factor charging recommendation system that turns fragmented information into actionable decisions
  • Designed a trust-aware crowdsourcing pipeline with geofence validation and cross-verification
  • Integrated AI receipt scanning to make charging records and carbon impact more visible and meaningful
  • Expanded the project beyond a basic station finder into a broader charging decision and sustainability platform

What we learned

  • In real-world AI products, trust and validation matter just as much as model capability
  • Prompt engineering combined with strong backend rules can outperform a rushed custom ML pipeline
  • Building for a problem you understand personally leads to sharper product choices
  • Transparency alone is not enough — users need help comparing options and making better decisions

What's next for ChargeSmart

  • Launch a WeChat Mini Program version to meet Chinese users where they already are
  • Expand from the current Shenzhen pilot to more cities such as Beijing and Shanghai
  • Improve recommendation quality with stronger personalization and station reliability scoring
  • Explore operator partnerships and data integration opportunities for more reliable live pricing
  • Continue developing the carbon feedback system so users can better understand the environmental impact of their charging behavior

Built With

Share this project:

Updates

Submission history