Anang AI

homepage

AnangAI

Making Kingston's tourism information accessible, sustainable, and intelligent.

Inspiration

AnangAI started from a need to make Kingston's tourism information easier to find and more accessible. Existing guides were scattered, hard to search, and often missed details like accessibility and sustainability. We wanted a single place where visitors and locals could ask questions in natural language and get accurate, up-to-date answers.

Our Focus:

Sustainable Tourism: Highlighting Green Plate–certified restaurants and local businesses
Accessibility: Including accessibility and washroom information for all places
Multilingual Support: Serving both English and French speakers
Intelligent Search: Using AI to understand natural language queries

What We Learned

Technical Skills

RAG (Retrieval Augmented Generation)
Built a system that retrieves relevant data from local files and augments LLM prompts for accurate, context-aware responses

Full-Stack Development
Integrated a React frontend with a FastAPI backend, handling state, routing, and API communication

*Data Processing & Web Scraping *
Scraped data from different sources using Beautiful Soup library and normalized data from multiple sources (restaurants, places, events) with varying formats

Multilingual Architecture
Implemented i18n with React Context and JSON translation files

Smart Query Processing
Built keyword mapping, date parsing, and query specificity detection

Design & UX

Created a ChatGPT-like interface with animated message bubbles and typing indicators
Used Framer Motion for smooth transitions
Designed responsive cards with consistent layouts
Implemented visual indicators (leaf icons for certifications, accessibility icons)

How We Built It

Architecture

Frontend (React + Vite)

React 19 with React Router for navigation
Tailwind CSS for styling
Framer Motion for animations
Context API for global state (language, chat history, admin auth)
Custom hooks for API calls and translations

Backend (FastAPI + Python)

FastAPI REST API with CORS middleware
RAG system using OpenRouter API (GPT-4o-mini)
Dynamic file discovery for Food, Places, and Events folders
Smart search with keyword mapping and date filtering
Data normalization for accessibility and washroom information

Key Features Implemented

Intelligent Chat Interface

Natural language processing for queries like "show me restaurants near water"
Date-based filtering for events ("events on Feb 8")
Certification-based prioritization (Gold > Silver > Bronze)
URL hiding behind "Find Location" links

Dynamic Discovery Page

Auto-generates navigation tabs from data files
Displays entries in responsive card layouts
Shows accessibility and washroom information with icons

Data Management

Parses entries from multiple file formats
Normalizes accessibility values (Full Access, Partial Access, Limited)
Handles NULL values and edge cases

Multilingual Support

English and French translations
Language toggle on all pages
LLM responses in the selected language

Local Business Registration

Help out local business to grow by making them visible more on our website
They can send request by filling all required info
The info will be sent to admin and after their approval, their business will be visible on this website

Challenges We Ran Into

1. Data Normalization

Problem: Data files had inconsistent formats—some entries had "NULL" (uppercase), others had "null" (lowercase), and accessibility descriptions varied.

Solution: Built a normalization function that:

Handles case-insensitive matching
Extracts key phrases from longer descriptions
Falls back to "null" when values can't be determined python # Example: "Partial Access - Historic buildings vary..." → "Partial Access" if "partial access" in accessibility_lower or "partial" in accessibility_lower: entry["accessibility"] = "Partial Access"

2. Query Specificity Detection

Problem: Distinguishing between vague queries ("show me some restaurants") and specific ones ("pub near Princess Street").

Solution: Built a keyword mapping system that:

Detects query intent (food, places, events)
Returns samples for vague queries
Returns full lists when requested
Filters by location and date when specified

3. Multilingual LLM Responses

Problem: Ensuring the LLM responds in the correct language with proper translations.

Solution: Added language-specific instructions to prompts:

Explicit translation mappings (Location → Emplacement)
Section header translations
Example formats in both languages

Accomplishments That We're Proud Of

RAG Architecture: Retrieves relevant context from 100+ entries across multiple categories
Smart Search: Handles typos, synonyms, and fuzzy matching
Date Parsing: Extracts dates from natural language ("events on feb 8")
Certification Prioritization: Sorts food entries by Green Plate Certification automatically
Accessibility First: Normalizes and displays accessibility information consistently