Golden Guide

The challenge

QHacks 2026 used “Golden Age” as its theme. We focused on a practical problem inside that brief: municipal services exist for seniors, but finding the right program, checking eligibility, and figuring out how to get there can become a project of its own.

Golden Guide was our attempt to make that process feel less like searching through tabs and more like asking one patient guide for help.

More than a chatbot

I did not want the demo to stop after returning a paragraph. Golden Guide uses an agentic loop with 11 tools so it can move from a question toward an action. It can search for services, check eligibility information, look up transit, draft an email, and offer a phone call after asking for confirmation.

That confirmation step mattered. A useful assistant should not turn a vague request into an external action without making the next step clear first.

Building around real local information

Transit

We used real Kingston Transit GTFS data for route and trip information. The goal was to connect a service recommendation to the practical question that often comes next: how do I actually get there?

Accessibility

The interface supports English and French and was designed around direct language, clear actions, and a lower-friction path through the result. We treated accessibility as part of the product flow rather than a final styling pass.

The stack

The build combined Gemini with a FastAPI backend and a Next.js interface. ElevenLabs, Twilio, and SendGrid were part of the broader toolset for voice and communication workflows.

The result

We built Golden Guide in 36 hours and won Best Use of Gemini at QHacks 2026.

The award was exciting, but the more useful lesson was architectural. Agentic systems feel convincing when each tool has a narrow job, the user can understand what is happening, and consequential actions pause for confirmation.