Explore-Singapore

Explore Singapore - Legal, Historical, and Infrastructural Knowledge Engine

The Explore Singapore project is an advanced intelligence engine designed to provide accurate and relevant information regarding Singapore's legal framework, policies, historical events, and critical infrastructure. This system employs a sophisticated Retrieval-Augmented Generation (RAG) approach to mitigate inaccuracies often found in other language models, ensuring that users receive data grounded in factual reality. Leveraging a comprehensive dataset comprising over 33,000 pages of curated documents, this platform aims to enhance the user's understanding of Singaporean law and governance.

Key Features

Triple-AI Failover Backend

To guarantee reliability, the system is powered by a robust chain of command for language model inference consisting of:

• Primary: Google Gemini 2.0 Flash
• Secondary: Llama 3.3 70B via OpenRouter
• Tertiary: Llama 3.3 70B via Groq
  This architecture provides a 99.9% uptime, making it ideal for heavy traffic scenarios.

Interactive "Liquid-Glass" User Interface

The user interface utilizes modern web design principles, featuring a custom-built React component with:

• Glassmorphism: Aesthetic backdrop blur for a sleek appearance.
• Smooth Animation: Spring physics for fluid interactions.
• Minimalist Design: Emphasis on clarity with SVG icons and elegant typography.

Local Embedding Inference

The platform prioritizes performance and privacy by executing the embedding model locally, avoiding external API calls that could introduce latency and additional costs.

System Architecture

The system architecture employs a high-performance RAG pipeline optimized for low-resource environments, comprising the following stages:

1. Ingestion: Processing of over 33,000 pages of legal and historical documents from Singapore.
2. Vectorization: Utilizing BGE-M3 to produce 1024-dimensional semantic embeddings.
3. Retrieval: Implementation of FAISS (Facebook AI Similarity Search) to achieve millisecond-latency vector lookups.
4. Generation: A failover logic that ensures continuous uptime and performance.

Tech Stack

Component	Technology	Description
Frontend	React, Framer Motion	Interactive "Ask AI" widget.
Backend	Flask, Gunicorn	REST API managing RAG logic.
Vector DB	FAISS (CPU)	Local, high-speed similarity search.
Embeddings	Sentence-Transformers	BGE-M3 (On-premise execution).
LLMs	Gemini 2.5 flash, Llama 3.3	Text generation and synthesis capabilities.
Deployment	Hugging Face Spaces	Cloud hosting using Docker technology.

Explore Singapore is not just a knowledge repository; it is a comprehensive tool designed to facilitate a greater understanding of Singapore's legal landscape, its policies, and the history that shapes this vibrant nation.