MobiClear

Inspiration

Mobility services in Vietnam are evolving rapidly — from ETC (Electronic Toll Collection) to smart transport infrastructure and automotive services.
But users still struggle with something very basic: understanding the rules.

People often don’t know why they were fined, how ETC actually works, what fee they’re being charged, or why their RFID card failed. These problems come from confusion, not technology.

We realized that mobility becomes smooth only when the rules are clear.
So we created MobiClear — a simple assistant that makes mobility regulations understandable for everyone.

What it does

MobiClear is an AI assistant that instantly explains anything related to:

• ETC lane rules & penalties
  
• How RFID ETC cards work
  
• Common mistakes and how to avoid them
  
• Toll fees, mobility notices, and service explanations
  
• Traffic signs, warnings, or documents through image upload
  
• Any confusing text, bill, or warning — explained simply
  

Key features:

• Chat to ask questions in natural Vietnamese
  
• Upload images of bills, traffic signs, warnings
  
• Instant explanations in clear, friendly language
  
• Built on a curated mobility knowledge base
  
• Ready for deployment into mobility platforms such as Tasco’s ecosystem
  

How we built it

We built MobiClear through four main components:

1. A curated Mobility Knowledge Base

We manually summarized and structured essential mobility information:

• ETC rules
  
• RFID card usage
  
• Common user errors
  
• Fee concepts and mobility guidelines
  
• Automotive & service ecosystem context
  All of this is stored in a JSON knowledge base for grounding the AI.

2. A Reasoning Layer powered by LLMs

We crafted a prompt‑engineering pipeline combining:

• A system prompt defining behavior and tone
  
• A dynamic user prompt injecting knowledge base content
  
• A domain‑aware response generator
  This ensures answers are clear, accurate, and non‑generic.

3. Vision + OCR Pipeline

Users can upload:

• ETC bills
  
• Traffic signs
  
• Notices or warnings
  The system extracts text, interprets it, and explains it as if a human expert were guiding them.

4. A sleek, fast demo UI

A minimal, friendly chat interface was built using React:

• Chat bubbles
  
• Image upload
  
• Loading feedback
  
• Smooth, real-time conversation

Challenges we ran into

• Condensing complex regulations into simple explanations without losing accuracy
  
• Ensuring the AI does not hallucinate outside the curated knowledge base
  
• Handling noisy images and inconsistent text from OCR
  
• Keeping everything lightweight enough to build within 36 hours
  
• Designing a UX that feels natural, not like a generic chatbot

Accomplishments that we're proud of

• Delivered a fully functional end‑to‑end assistant
  
• Combined vision, knowledge grounding, and conversational AI in a single flow
  
• Created a solution that solves a real user pain point
  
• Built a UI that anyone can understand in 5 seconds
  
• Designed an idea that is practical and deployable, not just a hackathon demo

What we learned

• Domain‑specific AI is dramatically more useful than general AI
  
• Simplicity wins — users want clarity, not complexity
  
• Good prompt engineering can be more important than heavy ML work
  
• Mobility rules are not intuitive, and explaining them clearly has real impact
  

What's next for MobiClear

• Add multi‑language support (English, Vietnamese)
  
• Expand coverage to all mobility services in Vietnam
  
• Integrate with mobile apps, kiosks, and customer service portals
  
• Add voice interaction for drivers
  
• Personalized recommendations based on frequent routes
  
• Real‑time alerts: low ETC balance, new penalties, updated rules
  
• Potential partnership with Tasco to scale to millions of users

Built With

• amazon-web-services
• brightdata
• exa
• openai
• python
• telegrambot
• tinyfish
• trae
• valsea
• zalooa