← Projects
Meeting Prep AI
Voice-triggered AI assistant that researches who you're meeting and briefs you before calls. Built at Google Gemini Hackathon 2026.
gemini · voice-ai · hackathon · function-calling
Voice-triggered AI assistant that researches who you’re meeting and briefs you before calls. Built solo at the Google Gemini Hackathon Singapore (January 2026).
GitHub Repository: github.com/phuaky/gemini-sg-hackathon
Live Demo: connected-briefing-agent
Features
- Voice-Activated Briefings - Say “prepare me for my next meeting” to trigger research
- Automatic Attendee Research - Searches the web for information about who you’re meeting
- Bio Card Generation - Compiles key facts into a digestible briefing
- Voice Output - Speaks the briefing back to you hands-free
- Email Follow-up - Suggests sending confirmation emails post-briefing
Technical Stack
- Gemini API - Function calling orchestration and LLM reasoning
- Serper API - Web research and search capabilities
- Google Cloud TTS - Text-to-speech voice output
- Web Speech API - Browser-based voice input
- Express.js - Backend server
- Cloud Run - Serverless hosting on Google Cloud
Architecture
"Prepare me for my next meeting"
↓
Voice Input (Web Speech API)
↓
Gemini Function Calling
↓
Research Attendee (Serper API)
↓
Generate Bio Card
↓
Voice Briefing (Google Cloud TTS)
↓
"Should I send a confirmation email?"Current Status
Working (40% of vision):
- Voice Output (TTS)
- Function Calling orchestration
- Web Research via Serper
Not Yet Implemented:
- Real Google Calendar OAuth (using mock data)
- Real Gmail OAuth (mock JSON)
- Bio Card image generation
- Proactive triggers
- Email sending
- Gemini Live bidirectional voice (fell back to Web Speech API)
The Vision
Not just meeting prep - a relationship operating system:
- Proactive calendar monitoring (“meeting in 30 mins, here’s your brief”)
- Tracks your goals across relationships over time
- Remembers what you committed to in past meetings
- Suggests strategic asks (“You’ve met Sarah 4x. Ask for the intro this time.”)
- Bidirectional voice via Gemini Live API
- Visual bio cards via image generation
Lessons Learned
Built for the Google Gemini Hackathon: 76 projects, 200 participants, 7 hours.
The Gap:
- Tools Used: 5 vs 25+ (winners used extensive Google ecosystem)
- Architecture: Single agent vs Multi-agent orchestration
- Demo: No video vs Full production video
- Novel Tech: API wrapper vs custom algorithms
Key Insight: Hackathons reward different things than product building. Knowing the game you’re playing matters.
Hackathon Details:
- Event: Gemini 3 Hackathon by Google DeepMind + 65labs
- Date: January 11, 2026
- Location: Singapore
- Result: Didn’t place (Tier B- in self-analysis)