Inspiration
The inspiration for Radio Mirchi came from the realization that propaganda and misinformation spread through seemingly innocent channels that sound reasonable at first glance. We wanted to create an interactive experience that would help people understand how subtle manipulation works in media, while also being genuinely fun to play.
The retro radio aesthetic was inspired by classic spy movies and the golden age of radio broadcasting, when information was controlled by a few powerful stations. We thought: what if you could be the secret agent infiltrating these broadcasts to wake people up?
What it does
Radio Mirchi is a browser-based stealth game where players take on the role of an underground agent infiltrating AI-generated radio propaganda broadcasts. The core gameplay involves:
- Real-time Dialogue: Players engage with AI radio hosts through text, trying to subtly introduce counter-arguments without being detected
- Persuasion Mechanics: Track how many listeners you're "awakening" with a dynamic persuasion meter
- Stealth Elements: Avoid detection by the AI security system while gradually changing minds
- Mission Variety: Each playthrough generates unique radio show scenarios with different topics and host personalities
- Authentic Retro Experience: Complete with vintage terminal interfaces, pixel art, and classic computing sounds
The goal is to get 80% of listeners to start questioning what they're hearing before the AI catches on and cuts your connection.
How we built it
Frontend Architecture
- React 18+ - Because we like our components fresh
- TailwindCSS - Making things look good without the headache
- react-rnd - For those draggable windows that feel so satisfying
- RetroUI - Pre-made retro components (because why reinvent the wheel?)
- react-sound-visualizer - Those cool waveforms you see
- Pixabay retro sounds - All the bleeps and bloops
Backend Architecture
- FastAPI as the core web framework for high-performance async operations
- WebSocket connections for real-time gameplay communication
- Google Gemini 2.5 Flash for generating convincing AI radio host dialogue
- Deepgram SDK for text-to-speech and speech-to-text processing
- MongoDB for mission data persistence and player progress tracking
- Motor (async MongoDB driver) for non-blocking database operations
The Two-Phase System
We implemented a unique two-phase architecture:
- Phase 1: Heavy AI processing happens upfront - generate mission context, create characters, establish scenarios
- Phase 2: Real-time streaming dialogue with minimal latency for smooth gameplay
Real-time Audio Pipeline
Player Voice → Web Speech API → Backend Processing → AI Response → Deepgram TTS → Audio Chunks → Frontend Playback
Challenges we ran into
Real-time Audio Streaming
The biggest challenge was making voice interactions feel natural and responsive. AI text generation takes time, but players expect instant responses. We solved this by:
- Streaming audio chunks as they're generated rather than waiting for complete responses
- Implementing a background processing system for heavy AI computations
- Using WebSocket connections for low-latency communication
AI Context Management
When players interrupt AI hosts mid-sentence, the system needed to remember exactly where it was and resume naturally. We implemented:
- Precise interruption point tracking
- Context preservation across dialogue turns
- Smooth conversation resumption after player input
Making Propaganda Subtle Yet Detectable
Creating AI-generated propaganda that felt realistic without being too obvious or too subtle was an art form. We fine-tuned prompts to generate:
- Believable radio host personalities
- Scientific-sounding but questionable claims
- Gradually escalating propaganda techniques
- Natural conversational flow that players could realistically interrupt
Accomplishments that we're proud of
Technical Achievements
- Innovative Two-Phase Design: Created a system that balances AI processing complexity with real-time responsiveness
- Sophisticated AI Integration: Developed prompts that generate convincing, varied radio show scenarios
Innovation in Educational Gaming
- Stealth Learning: Players learn about propaganda techniques through gameplay rather than lectures
- Critical Thinking Development: Encourages players to question information sources and recognize manipulation
- Interactive Media Literacy: Demonstrates how seemingly innocent content can contain hidden agendas
What we learned
Technical Lessons
- Async Python is Powerful: FastAPI's async capabilities were crucial for handling multiple real-time connections
- WebSocket Planning is Critical: Real-time connections require extensive error handling and state management
- AI Prompt Engineering is an Art: Getting consistent, high-quality AI responses requires careful prompt crafting
- Audio Streaming Complexity: Real-time audio processing has many edge cases and requires robust buffering strategies
Architecture Insights
- Separate Heavy and Light Operations: Doing AI-intensive work upfront and keeping real-time operations simple
- State Management Matters: Clear separation between persistent data (MongoDB) and session data (memory)
- Error Handling is 80% of the Code: Happy path is easy; handling all the ways things can break is the real work
Game Design Realizations
- Subtle Mechanics Work: Players engage more when they discover game mechanics organically
- Authentic Aesthetics Matter: The retro styling wasn't just decoration - it enhanced immersion significantly
What's next for Radio Mirchi
Immediate Technical Improvements
- Voice Recognition Enhancement: Improve speech-to-text accuracy for better player expression recognition
- AI Host Personality Expansion: Add more diverse character archetypes and regional speech patterns
- Mobile Optimization: Adapt the interface for touch devices while maintaining the retro aesthetic
- Performance Optimization: Reduce latency further and optimize for lower-end devices
Gameplay Expansion
- Mission Variety: Add different types of media (TV shows, podcasts, social media feeds)
- Cooperative Mode: Allow multiple players to work together on the same infiltration mission
- Difficulty Scaling: Implement adaptive difficulty based on player performance
- Achievement System: Add progression rewards for different infiltration techniques
Educational Integration
- Analysis Mode: Post-game breakdowns showing propaganda techniques that were used
- Historical Scenarios: Missions based on real propaganda campaigns from history
Radio Mirchi represents just the beginning of what's possible when you combine gaming, AI, and education to tackle real-world problems in an engaging, accessible way.
Log in or sign up for Devpost to join the conversation.