-
Tell me what you're craving, and I'll find the perfect restaurant for you!
Inspiration
The idea for VibeEating.ai was born from a universal frustration: deciding “what to eat” is never as simple as it should be. Traditional food apps bombard you with star ratings, generic lists, and endless scrolling. They ignore the most important part — your current vibe.We’ve all typed “spicy Indian non-veg food tonight” into Google and still ended up disappointed. Or tried to find a place that nails a specific dish (the legendary butter chicken or the perfect pad thai) only to discover the restaurant’s overall rating doesn’t reflect how incredible that one item is.VibeEating.ai was inspired by the simple desire to have a smart foodie friend who truly understands your craving, dietary needs, budget, mood, and location — then instantly finds the places where real people are raving about the exact dish you want. We wanted to build the first app that treats food discovery like a natural conversation instead of a search form.
What it does?
VibeEating.ai is a conversational AI-powered chatbot (mobile + web) that turns casual free-text messages into hyper-personalized restaurant recommendations in seconds.How it works in real life:You chat naturally: “I want spicy Indian non-veg food tonight” or “Thai vegan drinks, not too spicy.” The AI instantly extracts your preferences (cuisine, dietary, spiciness, food/drinks, vibe). It gracefully asks for any missing details (location, budget, dine-in vs delivery, time, party size). It searches Google Places in real time + augments with Foodpanda/Grab delivery data. The magic: It scans actual reviews to detect dish-level raves (e.g., “butter chicken is legendary here” even if the restaurant rating is only 3.8). It returns the top 3–5 perfect matches with:Highlighted dish quotes from real reviews Clear dine-in vs delivery cost & time comparison table “Why this matches your vibe” transparent explanation Action buttons (View on Maps, Order on Foodpanda/Grab, Reserve)
It supports multi-turn conversation with full memory, corrections (“actually make it less spicy”), and ends gracefully. Recommendations are delivered in under 3 seconds.
How we built it
We followed a modern, scalable, full-stack architecture:Frontend: React Native (Expo) with Gifted Chat for the beautiful conversational UI + Web PWA support. Backend: Python + FastAPI for lightning-fast API endpoints. AI Brain: LangChain orchestration + Grok API (xAI) for preference parsing, intent detection, dish-level review analysis, and natural conversation flow. Database & Caching: PostgreSQL for user profiles/favorites + Redis for session memory and API caching (30–60 min TTL). Data Sources:Google Places API (New) with Text Search, Nearby Search, Place Details, and businessMenus field mask. Ethical Apify actors for real-time Foodpanda and Grab menu prices, fees, and ETAs.
Key Custom Components: Advanced system prompt for strict JSON preference extraction + proactive follow-ups. Second Grok call for dish-rave sentiment analysis on up to 5+ reviews per restaurant. Smart ranking engine: vibe relevance + dish sentiment > distance > rating > price.
The entire data flow (message → parse → search → analyze → enrich → respond) was optimized for speed and reliability.
Challenges we ran into
Real-time multi-source data synchronization: Google Places and delivery platforms don’t share official unified APIs, so we had to ethically combine Google data with Apify scrapers while respecting rate limits and robots.txt. Dish-level intelligence accuracy: Teaching an LLM to reliably extract specific dish mentions and sentiment from unstructured reviews (without hallucinating) required heavy prompt engineering and multiple iteration cycles. Conversational memory & corrections: Maintaining perfect context across multi-turn chats while allowing users to change their mind mid-conversation was tricky; Redis session management took several refinements. Performance under 3 seconds: Balancing LLM calls, Google API latency, and review analysis while keeping the experience snappy required aggressive caching and parallel processing. Privacy & compliance: Handling sensitive dietary data without storing it permanently (GDPR/CCPA) meant careful session-only processing and clear consent flows.
Accomplishments that we're proud of
Built a true dish-level recommendation engine that goes far beyond star ratings — this is the feature users get most excited about in demos. Achieved sub-3-second end-to-end recommendations even with multiple LLM calls and external API integrations. Created a chatbot experience that feels genuinely delightful and human — warm, witty, and never robotic. Successfully integrated Google Places + Foodpanda/Grab data into clean, side-by-side cost comparison cards. Delivered full session memory, proactive follow-ups, and graceful error handling (no-results → smart alternatives). Launched a polished project website (https://app-b2izzanrot1d.appmedo.com/) with the complete “About the Project” page live.
What we learned
Prompt engineering is 70% of LLM-powered product success — small changes in system prompts dramatically improved JSON adherence and conversation quality. Real-world food data is messy; combining structured APIs with review scraping taught us the importance of robust fallback logic and ethical data practices. Users value transparency (“Why this matches your vibe”) more than we expected — explaining the AI reasoning builds instant trust. Speed and delight matter more than feature count. A fast, joyful chat experience beats a feature-heavy app every time. Cross-platform development (React Native + FastAPI) with serverless-friendly design makes scaling to 10k+ daily users feel achievable from day one.
What's next for VibeEating.ai
Mobile App Launch: Full iOS & Android native apps in the next 4–6 weeks (currently in private beta). Personalization Engine: User accounts + long-term memory so the AI learns your favorite spots and past vibes over time. Push Notifications: “Your favorite Thai spot just added a new spicy vegan special!” and smart re-engagement. Multi-language Support: Full regional language detection and responses. Monetization: Affiliate links to delivery platforms + optional “Pro” tier for even faster/more advanced suggestions. Expanded Data Sources: More cities, more delivery platforms, and optional voice input. Community Features: Save favorites, share recommendations, and “vibe matching” with friends.
Log in or sign up for Devpost to join the conversation.