🍽️ Austin Eats — AI Restaurant Search

AI-powered restaurant discovery app built using Google Gemini + AlloyDB. This project enables users to search restaurants using natural language or semantic search, powered by embeddings.

---

Live Demo

• App: Live App Link
• Demo Video: Watch on Google Drive

---

Features

Natural Language Search

• Users can type queries like "cheap tacos" or "best Japanese restaurants" or "high rating restaurants".
• Google Gemini converts the query into SQL.
• SQL is executed on AlloyDB in real time.

Semantic Search (Vector Search)

• Uses embeddings (text-embedding-005).
• Finds restaurants based on meaning, not just exact keywords.
• Returns a similarity score for each result.

Explainable AI

• Displays generated SQL queries directly in the UI.
• Improves transparency and simplifies debugging.

Modern UI

• Clean, responsive interface.
• Dynamic result cards.
• Interactive search suggestions.

---

Architecture

System Flow

Frontend (HTML + JS) → Flask Backend (API) → Gemini API → SQL Generation → AlloyDB (PostgreSQL) → Results returned to UI

Use Case Diagram

---

Screenshots

---

Tech Stack

• Frontend: HTML, CSS, JavaScript
• Backend: Flask (Python)
• Database: AlloyDB (PostgreSQL)
• AI Models: Gemini (NL → SQL), text-embedding-005 (semantic search)
• ORM / DB: SQLAlchemy
• Deployment: Google Cloud Run
• Containerization: Docker

---

📁 Project Structure

.
├── app.py
├── templates/
│   └── index.html
├── assets/
│   ├── usecase_diagram.png
│   └── screenshots/
│       ├── image_2026-03-27_17-26-45.png
│       ├── image_2026-03-27_17-36-25.png
│       └── image_2026-03-27_17-21-25.png
├── requirements.txt
├── Dockerfile
└── README.md

---

Cloud Deployment (Google Cloud Run)

The application is designed to run in Google Cloud due to AlloyDB private networking and Gemini API integration. Example deployment command:

gcloud beta run deploy restaurant-ai-finder \
  --source . \
  --region=us-central1 \
  --network=easy-alloydb-vpc \
  --subnet=easy-alloydb-subnet \
  --vpc-egress=all-traffic \
  --allow-unauthenticated \
  --clear-base-image \
  --set-env-vars GEMINI_API_KEY='YOUR_GEMINI_API_KEY',DATABASE_URL='YOUR_ALLOYDB_CONNECTION_STRING'

---

API Endpoints

• Get All Restaurants: GET /api/restaurants
• Natural Language Search: GET /api/search?query=your_query
• Semantic Search: GET /api/semantic-search?query=your_query
• Health Check: GET /health

---

Safety Considerations

• SQL queries are generated via AI → must be validated before execution.
• Only SELECT queries should be allowed (recommended improvement).

---

Future Improvements

1. Add query validation for security
2. Add user feedback (thumbs up / down)
3. Improve UI with loading skeletons
4. Add location-based filtering
5. Cache Gemini responses for performance

---

Hackathon Context

Built as part of Google Gen AI Academy APAC — Track 3 (Cohort-1)

Focus: - AI-ready databases

• Natural language data access
• AlloyDB + Gemini integration