Singaporean Developer Builds AI Legal Assistant Using RAG Technology

November 11, 2025

Singapore – In a project that blends civic-minded innovation with cutting-edge artificial intelligence, a developer has independently built a specialized large language model (LLM) designed to navigate the complex landscape of Singaporean statutes and public policies. The system, named "Explore Singapore," utilizes a technique called Retrieval-Augmented Generation (RAG) to ground its responses directly in official government documents, aiming to provide citizens, developers, and travelers with reliable legal information.

The term "built", as defined by Merriam-Webster, refers to something that has been formed or constructed. In this context, the developer has constructed a sophisticated digital tool from the ground up. The project's core objective was to create a domain-specific intelligence engine that could decrease errors common in generic LLMs by using Singapore's public policy documents and legal statutes as its exclusive information source.

Bridging the Gap Between AI and Accurate Legal Information

The motivation behind the project stems from a common frustration: the difficulty of parsing lengthy, complex government PDFs. "The objective required building a domain-specific search engine which enables LLM systems to decrease errors by using government documents as their exclusive information source," the developer, who goes by the online alias Fantastic_suit143, explained in a project announcement.

The system ingests a massive corpus of 594 PDFs concerning Singaporean laws and acts, comprising roughly 33,000 pages. Using Python scripts and the PyPDF2 library, the documents are parsed, chunked into manageable text segments, and converted into numerical vectors using the Hugging Face BGE-M3 embedding model. These vectors are stored in a FAISS database for rapid similarity search, orchestrated by the LangChain framework.

How RAG Ensures Accuracy

The power of the system lies in its RAG pipeline. When a user submits a query—such as "Can I fly a drone in a public park?"—the system does not rely on the LLM's internal, potentially outdated or generalized knowledge. Instead, it first retrieves the most relevant text chunks from its verified document database. These chunks are then synthesized into a coherent answer by the LLM, complete with citations.

The developer demonstrated the stark difference this makes. A standard LLM provided generic advice about "checking local laws." In contrast, the RAG-powered system specifically cited Singapore's Air Navigation Act, detailed 5km no-fly zones, and linked directly to the Civil Aviation Authority of Singapore (CAAS) permit page. "The difference was clear and it was sure that the AI was not hallucinating," the developer noted.

Technical Architecture and Challenges

The tech stack is a testament to modern, accessible AI development tools. The backend is powered by Flask, with a React and Framer frontend. For resilience, the system employs a primary LLM (Google's Gemini) with a backup model (Arcee AI's Trinity Large) to ensure uptime, with tailored system instructions for each to maintain response quality.

However, building such a system is not without its hurdles. The developer is currently focused on optimizing the ranking strategy within the RAG architecture. A common challenge in RAG systems is the retrieval of irrelevant documents, which can skew or confuse the final answer. The project employs a multi-query retrieval strategy, breaking down complex questions into keywords to improve document matching.

"It's still in the development phase but still it provides near accurate information," the developer acknowledged, inviting feedback from the community to refine the platform further. The entire project is open-source and available on GitHub, encouraging collaboration and scrutiny.

Implications for Public Access to Law

Projects like Explore Singapore highlight a growing trend of using AI to demystify legal and bureaucratic information. For Singaporeans, businesses, and visitors, such a tool could significantly lower the barrier to understanding regulations on everything from housing grants to business licensing.

While not a substitute for professional legal counsel, it serves as a powerful first-stop resource. The act of building this tool—much like the construction platforms referenced by sources like Built Technologies, which focuses on synchronizing real estate finance—involves creating a structured, reliable system from disparate parts. In this case, the parts are thousands of pages of legal text, synthesized into an accessible interface.

The developer's work underscores how individual initiative, combined with robust open-source AI tools, can create public goods that address specific civic needs. As RAG technology continues to mature, its application in legal, educational, and governmental domains promises to make authoritative information more transparent and readily available than ever before.