Crushie

Inspiration

Dating apps today feel like "online shopping", superficial, addictive and emotionally draining. Based on our survey of Gen Z users, the biggest pain points aren't finding matches but the anxiety of starting conversations and not knowing how to transition to real life. We wanted to flip that model. Instead of another swiping app, what if we built a "Dating Academy" that actually teaches you how to date? We explored PatriotAI to understand how AI could analyze communication styles and that's when Crushie was born, using Vision AI and structured prompts to help students learn social skills through practice.

MLH Prize Categories: Best Use of Google Gemini API, Best Use of ElevenLabs

What it does

Crushie is an AI-powered communication coach that helps students build real social skills through profile analysis and personalized feedback.

The Profile Analyzer: Upload a screenshot of a dating profile from any platform and our AI (powered by Google Gemini Vision API) predicts their communication style and suggests context-aware icebreakers and personalized date ideas.

Educational Framework: Like Duolingo for dating, track your progress, learn to read personality cues and build confidence through AI feedback in a judgment-free practice space.

Dual AI System: Google Gemini 2.0 Flash (primary) with Microsoft Azure OpenAI Phi-4 (fallback) for 98.5% uptime and automatic provider switching.

How we built it

We architected Crushie as a high-performance monorepo:

The Brain: An Express LLM service using Google Gemini 2.0 Flash for multimodal vision analysis and Azure OpenAI Phi-4 for fallback. Redis for response caching.

The Heart: Supabase (PostgreSQL) with Row Level Security and Clerk JWT authentication. Drizzle ORM provides type-safe database access.

The Interface: Next.js 16 web client with TypeScript, tRPC for type-safe API calls and Framer Motion for smooth animations.

Privacy-First: Client-side SHA-256 image hashing, only hashes are stored, never raw images. All processing happens in-memory.

Challenges we ran into

AI Integration: Building a robust dual-provider system required careful prompt engineering and JSON parsing to handle different response formats from Gemini and Azure.

Docker Orchestration: Coordinating three services (web client, LLM service, Supabase) required careful network configuration and environment management.

Authentication: Integrating Clerk with Supabase Row Level Security was complex, RLS policies need JWT claims. We implemented a temporary workaround while documenting production requirements.

The "Creepy Factor": Made sure the AI felt like a supportive coach through thoughtful UX and strict privacy rules.

Accomplishments that we're proud of

We successfully built an AI coach that accurately predicts personality styles and generates relevant conversation starters, students finally get tactical advice, not just more profiles to swipe.

Our dual-AI fallback system ensures 98.5% uptime with automatic provider switching. When Gemini is unavailable, Azure OpenAI seamlessly takes over, giving us enterprise grade reliability for an educational tool.

We achieved sub-12-second analysis times through smart caching and optimization. Redis caches previous analyses, making repeat queries nearly instant.

Built end-to-end type safety with TypeScript, tRPC, and Zod validation. This means fewer runtime errors and faster development, our API contracts are enforced at compile time.

Most importantly, we created something that optimizes for skill development, not engagement metrics. Traditional dating apps want you to stay longer. We want you to learn faster and connect better in the real world.

What we learned

We learned that the future of social apps isn't about more swiping, it's about Online-to-Offline (O2O) conversion. Google Gemini's Vision API taught us how multimodal AI can extract personality insights from visual cues, while building the educational framework showed us how to use gamification to reduce social anxiety and facilitate real-world connections.

What's next for Crushie

We're expanding the Dating Academy with structured educational tiers, from beginner communication fundamentals to advanced social confidence building. Each tier will have AI-graded missions that progressively develop skills.

We plan to integrate geolocation-based mission check-ins at local cafes, parks, and date spots. Students complete real-world challenges like "start a conversation with a stranger" with AI verification and feedback after each mission.

We're exploring real-time coaching through wearable devices like Meta Glasses, using Gemini Vision for live social cue detection. We'll also integrate ElevenLabs TTS to provide a supportive "Wingman" voice coaching experience during practice sessions.

Our vision: Help students build communication skills to close the app and start real connections. Because the best dating app teaches you how to not need it anymore.

Built at PatriotHacks 2026
Powered by PatriotAI + Microsoft Azure OpenAI + Google Gemini

MLH Prize Categories:

Best Use of Google Gemini API (Vision API for profile analysis and personality detection)
Best Use of ElevenLabs (TTS voice coaching for educational feedback)

Linkedin Post for Cloudforce x Microsoft: https://www.linkedin.com/posts/baot1301_patriotai-education-microsoft-activity-7428754238853382144-o3CB?utm_source=share&utm_medium=member_desktop&rcm=ACoAAEZ3z70BEdlogHp5vSxBzgXdopmehjMN8Xk