YesOrNext

thumbnail
tagline
the problem
key insight
system architecture
thumbnail v2
user flow
design progress

Inspiration

We’ve all been there: you open Yelp, see 40 restaurant options, scroll for 20 minutes, and still can’t decide. The problem isn’t finding restaurants—it’s making a decision.

Current apps are built around lists and filters, not around helping users actually choose. We were inspired by the simplicity of dating apps like Tinder, where you make quick decisions one at a time. What if we applied that same “Yes or Next?” approach to restaurant discovery?

What it does

Yes or Next is a conversational, AI-powered restaurant discovery app that never shows a giant list. Instead, it presents one carefully curated restaurant recommendation at a time and asks: “Yes or Next?”

Key features:

Dual modes: Chat (text) or Talk (voice) — speak or type your cravings
One-at-a-time: Swipeable restaurant cards (Tinder-style) — no overwhelming lists
AI learning: Each swipe refines recommendations based on your preferences
Complete flow: From query → recommendation → selection → reservation/handoff → tracking
Voice integration: Full voice pipeline using OpenAI Whisper (speech-to-text) and TTS (text-to-speech)

How we built it

Frontend: Next.js 16 + React 19 + TypeScript + Tailwind CSS 4

Swipeable card interface
Dual-mode UI (Chat/Talk)
Real-time conversational experience
Reservation UI + “lock it in” handoff

Backend: FastAPI + Python 3.11

Async REST API
Yelp AI Chat API v2 integration
OpenAI Whisper for speech-to-text
OpenAI TTS for text-to-speech

Database: Supabase (PostgreSQL)

Conversation/session tracking
User preference signals from swipes
Restaurant caching
Reservation records

Infrastructure: Docker + Google Cloud Run

Containerized deployment
Scalable API services

APIs: Yelp AI, OpenAI Whisper, OpenAI TTS

Challenges we ran into

Voice processing pipeline: Integrating browser MediaRecorder with Whisper and handling audio formats reliably
State management: Keeping session context consistent across multiple pages and modes (Chat ↔ Talk ↔ Option)
Yelp AI integration: Understanding the response structure, handling rate limits, and feeding back “Next/Yes” context
Swipe UX: Creating smooth, native-feeling gestures and animations without jank
Conversation continuity: Maintaining the same “brain” across turns so it actually improves after each swipe
Cross-mode navigation: Letting users switch modes without losing the thread or recommendation state

Accomplishments that we’re proud of

Full Voice + Text dual mode — seamless mode switching with natural voice input/output
Decision-first UX — eliminated choice paralysis with binary “Yes/Next” decisions
Intelligent learning loop — swipes become feedback that sharpens the next recommendation
End-to-end flow — from request → recommendation → reservation/handoff → tracking
Production-ready architecture — type-safe, scalable, Dockerized, cloud deployed
Polished UI — modern design with smooth swipe interactions
Robust handling — graceful failures and user-friendly errors (including API limits)

What we learned

AI API integration: Prompting + structured parsing matter as much as the model—guardrails make the UX reliable
Voice UX is product UX: Small details (record/stop timing, latency, audio formats) make or break the experience
Decision-first design works: Reducing cognitive load with binary choices creates faster, happier outcomes
Full-stack contracts help speed: TypeScript + Pydantic-style validation makes iteration safer under hackathon time
Hackathon strategy: Protect the core loop first (“ask → one card → swipe → improve”) before adding polish

What’s next for YesOrNext

Short-term:

Enhanced personalization and preference learning (“safe bet” vs “try something new”)
Wishlist feature with AI re-ranking (“Tonight: pick from my list”)
Deeper booking flow (availability + direct reservation where possible)
Social sharing and saved history

Long-term: