Ragunauth RamsaroopPublic-value intelligence
Local intelligence layer

RR Smart Guide

Search the public site, identify the likely decision context, receive source-linked recommendations, and optionally run semantic or local AI modes in your browser.

No API key requiredLocal learningSource-linkedOptional on-device AI

Ask the site

Use plain language. Examples: “How do I assess mining carbon?”, “What do I have on Guyana investment?”, or “Help me prepare for a government meeting.”

Intent: waiting for your query

Semantic mode loads a small embedding model only when you select it.

Optional local LLM

Source-bounded local answer

This mode uses WebGPU and WebLLM to run a small language model inside your browser. The first run downloads model files. The answer is grounded only in site sources selected by the Smart Guide.

Nothing downloads until you select the button.

How the smart layer works

1. Understand intent

The worker maps your query and current page to themes such as mining, energy, water, government, ESG, community, executive leadership and research.

2. Retrieve evidence

MiniSearch ranks the public index locally. A Web Worker keeps heavier processing off the page thread.

3. Adapt locally

IndexedDB stores a short history on this device so the site can recommend a more relevant next resource. No page history is sent to the server by this feature.

Recent local learning

Loading local history…

Open-source components reviewed

The smart layer uses locally vendored, audited browser libraries from GitHub: MiniSearch for full-text retrieval, Comlink for Web Worker communication, and idb-keyval for local IndexedDB memory. Optional AI modes load Hugging Face Transformers.js or MLC WebLLM only after a user selects them.

Review methodology →