DealMate - Swipe deals with Gemini 3 Pro

🚀 Inspiration: The "Social Friction" Tax

We noticed a recurring problem in our office: promotional deals like "Buy 2 Get 1" were going to waste. People weren't losing money because they were rich; they were losing it because asking a colleague to split a deal felt socially awkward.

The "aha!" moment? Seeing a colleague buy two bubble teas, throw one away, and walk off—simply because they were too shy to ask for a partner. We realized that Gemini 3 could act as the ultimate social buffer, orchestrating complex, real-world workflows to save money and reduce waste.

🛠️ What DealMate Does

DealMate is an autonomous office marketplace powered by three intelligent modes:

1. The Deal Matcher ("Tinder" for Office Savings)

• How it works: Users swipe through nearby promos.
• The AI Brain: Gemini 3 analyzes preferences, schedules, and locations.
• The Result: When you swipe right, the AI instantly pairs you with a compatible colleague and handles the coordination via notifications. No awkward asking required.

2. The Walking Hero System

This system turns a coffee run into a community service. Gemini assigns "Heroes" based on:

• Path Efficiency: Who is already walking toward the shop?
• Schedule Check: Does the user have a meeting in the next 15 minutes?
• Reliability: Historical ratings of the volunteer.
• Rewards: Heroes earn points, badges, or even free items for their trouble.

3. The Deal Creator (Multimodal Vision)

• Input: A quick photo of a messy, handwritten store poster.
• Processing: Gemini Vision API parses the text, understands the context (e.g., "BOGO" or "30% off after 5 PM").
• Output: Structured, digital deal cards published to the entire building instantly.

💻 Tech Stack & Architecture

🚧 Challenges We Overcame

• Mobile Gesture Precision: React’s synthetic events struggled with swipe velocity. We implemented @use-gesture to ensure the "Tinder" feel was snappy and responsive.
• Structured Output: Early prompts returned inconsistent JSON. We refined our prompt engineering to enforce strict schemas, ensuring user IDs weren't hallucinated.
• Latency vs. UX: Gemini API calls (500-1500ms) can lag during rapid swiping. > Our Fix: We implemented Optimistic UI updates, allowing the card to fly off-screen immediately while the AI processes the match in the background.

📈 Future Scalability

To optimize the "Walking Hero" pathing, we could even model the efficiency gain mathematically. If is the distance to the shop and is the number of participants:

This ensures that the "Hero" is always taking the most logical route for the group.

Built With

• framer
• gemini
• motion
• react
• tailwind