GIC Demo
A prototype showing what a better Government Information Centre (GIC) could look like — specifically, Sri Lanka’s GIC, which is nominally meant to help citizens find government services and information but is, in practice, not great at it.
The demo is a single HTML file that loads a collection of markdown documents locally and uses Claude (Anthropic’s API) to provide a RAG (Retrieval Augmented Generation) search interface on top of them. For comparison, it also includes a plain search variant that works without any LLM.
What’s in it
- 623 refined documents rewritten according to content guidelines for clarity, plain language, and information completeness
- 624 original documents for comparison
- 3 content guidelines covering writing standards, visual design, and information management
- rewriter_lite.py — Python script that transforms documents using the content guidelines
- validator.py — quality assurance processing
The rewriting was done with Claude, guided by content guidelines I wrote that push for plain language, clear structure, and actually useful information (as opposed to the bureaucratic boilerplate that government sites tend to accumulate).
How to run it
- Download the repository
- Open
GIC - llm search.html - Replace
API-key-herewith your Anthropic API key - Open in a browser
This is a prototype for showcase purposes only. The documents are not fully vetted for accuracy and this is not a production-ready system of any kind.
Why this exists
Sri Lanka’s public information infrastructure is genuinely difficult to navigate. This was built to demonstrate that a RAG-based approach on top of well-structured content can dramatically improve findability — and to make the case that the bottleneck is content quality as much as technology.