🚘 DashGem – AI Car Dashboard Assistant

DashGem is an AI-powered app that helps drivers understand warning lights on their car's dashboard by snapping or uploading a photo. Powered by Gemini 1.5 Pro, DashGem delivers accurate, grounded, and chat-style explanations with actionable next steps.

You can find our demos here:

• Android (Kotlin) Version https://www.youtube.com/watch?v=hJiWXm2iC38
• IOS & Web (Flutter) Version https://www.youtube.com/watch?v=yGHNkooMWwQ

---

🛠️ Built with:

• 🤖 Kotlin (Jetpack Compose) for Android (uses Gemini SDK directly)
• 📱 Flutter for Web & iOS (connects to the Node.js backend)
• 🌐 Node.js (Express) backend with Gemini API (used by Web & iOS)

✨ Features

• 📸 Dashboard Photo Input

  • Upload or capture a dashboard photo from your device
  • AI analyzes the lights and symbols shown

• 🧠 Gemini 2.0 Flash Integration

  • Android uses the Gemini SDK directly
  • Flutter Web/iOS sends requests to a Node.js server which forwards to Gemini

• 💬 Conversational Chat UI

  • Interact naturally with multi-turn conversations
  • The AI remembers previous messages and attached images

• 🖼️ Persistent Image Context

  • The dashboard image stays available throughout the session, enriching follow-up questions

• 🛠️ Mechanic-Inspired Prompting

  • Custom Gemini system prompt instructs AI to behave like a car diagnostics expert
  • Grounded, clear, non-technical responses

🧱 Architecture & Data Flow

Android (Kotlin)

• Built natively using Jetpack Compose
• Uses the Gemini Generative AI SDK directly from the device
• Compresses and sends dashboard image + user message in each prompt
• Secure API key setup via local.properties and BuildConfig

Flutter (Web & iOS)

• Frontend built with Flutter
• Sends multipart POST requests to the Node.js server
• Includes optional dashboard image and user question
• Uses animated chat bubbles, image previews, and a custom chat layout

Node.js Server (Express)

• Handles CORS, uploads (via multer), and Gemini API calls
• Validates and detects MIME type with mmmagic
• Sends prompt + optional image to Gemini 2.0 Flash
• Returns structured AI response to the Flutter frontend

🌐 Backend API (Node.js + Gemini)

The backend server handles requests from the Flutter Web/iOS frontend, processes dashboard images and user messages, and returns AI-powered responses.

🔁 Endpoint

POST /analyzeDashboardPic

Request (multipart/form-data):

• image (optional): a JPEG or PNG image of the dashboard
• text: the user's message (combined with memory of prior chat)

Example request from Flutter:

POST heroku-link
Content-Type: multipart/form-data

Successful Response:

{
  "response": "This indicator usually means low tire pressure. Please check all tires."
}

📍 Hosting

• ✅ Hosted on Heroku
• 🌐 Public endpoint used by Flutter Web/iOS:

  https://dash-gem-ef3cd0583e98.herokuapp.com/analyzeDashboardPic

• ⚠️ Note: GCP hosting setup is in progress and currently non-functional.

🧪 Running DashGem Locally

You can run DashGem's backend, Android app, and Flutter frontend separately.

---

▶️ Android App (Kotlin)

✅ Requires Android Studio

1. Open the android/ project in Android Studio.
2. In local.properties, add your Gemini API key:

geminiApiKey=YOUR_API_KEY

3. Run the app on an emulator or physical device.

---

▶️ Backend Server (Node.js)

✅ Requires Node.js v18+ and Heroku CLI (optional)

1. From the server/ folder:

npm install

2. Add your API key to a .env file:

GEMINI_API_KEY=your_key_here

3. Start the server locally:

node app.js

4. The API will be live at:

http://localhost:8080/analyzeDashboardPic

To deploy to Heroku:

heroku login
heroku create
git push heroku main

---

▶️ Flutter Web/iOS App

✅ Requires Flutter 3.x with web/iOS enabled

1. Navigate to the web/ project folder:

cd web/
flutter pub get

2. Make sure the server URL in code points to Heroku:

const String kBackendUrl = 'https://dash-gem-ef3cd0583e98.herokuapp.com/analyzeDashboardPic';

3. Run the app:

flutter run -d chrome

4. For iOS (macOS only):

flutter run -d ios

🧠 Gemini AI System Prompt

DashGem uses a custom-crafted prompt to guide the Gemini model's behavior and tone.

---

🔧 Prompt Purpose

The AI is instructed to:

• Act like a professional car mechanic
• Avoid off-topic, hallucinated, or fictional responses
• Be helpful, accurate, and easy to understand
• Remember prior user messages and any attached images

---

📝 Prompt Snippet

You are a highly skilled car mechanic who interprets icons on car dashboards.
Your task is to analyze the dashboard lights, identify possible issues,
and explain what each illuminated icon means. Be brief, accurate, and clear.
You will always remember past user messages and images. Avoid going off-topic.

This ensures that users get consistent, grounded explanations — even across multiple messages in a single chat session.

🚀 Roadmap & Future Enhancements

Here are some ideas and planned upgrades for upcoming versions:

• 🔎 OCR (Optical Character Recognition)

  • Automatically detect and extract dashboard text and icons from photos

• 🌐 Multilingual Support

  • Translate responses for non-English-speaking users

• 📶 Offline / Fallback Mode

  • Offer limited local guidance when the user has no internet

• 📈 Analytics Dashboard

  • Track the most common dashboard warning lights across users

• 📱 Push Notifications (Mobile)

  • Notify users of unresolved warnings or follow-up suggestions

• 💾 Local Chat History

  • Let users revisit previous sessions with persistent storage

---

We’re always open to ideas! Feel free to open an issue or discussion to suggest new features or improvements. 🚗✨