About the project
Inspiration
We were inspired by the frustration of poor WiFi experiences - everyone has been there: video calls dropping, games lagging, or streaming buffering at the worst moments. Traditional WiFi diagnostic tools are either too technical (showing raw dBm values) or too simplistic (just "good" or "bad"). We wanted to create something that could not only analyze WiFi quality but explain it in plain English, like having a tech-savvy friend who could tell you exactly why your connection is acting up and what to do about it.
What it does
hifi-wifi is a smart WiFi assistant that analyzes your connection quality and provides conversational explanations for optimization recommendations. The Android app collects WiFi measurements (signal strength, latency, bandwidth) and classifies them into user-friendly categories like "excellent" or "poor". It then makes algorithmic decisions about whether to stay put, move locations, or switch networks. The Raspberry Pi runs a local AI model (Qwen 3 0.6B) that explains these recommendations in natural, friendly language - no technical jargon, just clear advice about what will improve your WiFi experience.
How we built it
We built this as a distributed system with three main components:
Android App: Collects real-time WiFi data using native Android APIs (WifiManager, network tests) and classifies measurements into categorical labels. Implements rule-based decision logic to determine optimal actions.
Raspberry Pi Backend: Runs a Flask REST API that receives classified measurements and recommendation decisions from the Android app. Integrates with Ollama to run the Qwen 3 0.6B model locally for generating conversational explanations.
AI Explanation Engine: Uses a carefully crafted system prompt to ensure the LLM explains recommendations in plain English, avoiding technical terms like "2.4GHz" or "dBm" and focusing on user experience benefits.
The system communicates via HTTP over the local WiFi network, making it simple and reliable.
Challenges we ran into
AI Accuracy Issues: Initially, we tried to have the LLM both classify measurements and make recommendations, but this led to inconsistent results (only 43.8% accuracy). We solved this by offloading classification and decision-making to deterministic algorithms on the Android side, letting the LLM focus on what it does best - generating natural explanations.
Technical Complexity: We initially explored Bluetooth Low Energy (BLE) for communication but ran into compatibility issues with macOS testing and realized HTTP was simpler since both devices would be on the same WiFi network anyway.
User Experience Design: Crafting the right system prompt was crucial - we needed the AI to be conversational and helpful without being technical. This required multiple iterations to get the tone and content just right.
Accomplishments that we're proud of
- 83.3% accuracy in recommendation explanations after architectural improvements
- Local AI processing - no data leaves the user's network, ensuring privacy
- Conversational interface - explanations read like talking to a knowledgeable friend
- Modular architecture - clean separation between data collection, decision logic, and explanation generation
- Cross-platform compatibility - works on Android and Raspberry Pi with HTTP communication
What we learned
We learned that AI is best when it does what it's good at - in our case, generating natural language explanations rather than making complex decisions. We also discovered that local processing can be both powerful and privacy-preserving, and that user experience is just as important as technical accuracy. The project taught us about the importance of iterative design and testing, especially when working with language models.
What's next for hifi-wifi
- Real-time monitoring: Continuous WiFi quality tracking with proactive recommendations
- Machine learning integration: Learn from user behavior patterns to improve recommendations
- Multi-room mapping: Create WiFi quality heatmaps of homes/offices
- Integration with smart home systems: Work with existing IoT devices for automated optimization
- Mobile app store deployment: Make the solution available to a broader audience
- Advanced analytics: Historical data tracking and trend analysis for WiFi performance
Log in or sign up for Devpost to join the conversation.