Inspiration
We wanted a weekend planner that actually browses the web like a person would, not one that pulls from a static database or makes things up. Most recommendation apps are either outdated or generic. We wanted something that finds real, live results.
What it does
You type in your location and Weekend Scout finds you a well-reviewed local restaurant and a nearby hiking trail, then calculates the driving distance between the two. Everything is scraped live from Yelp, AllTrails, and Google Maps.
How we built it
We used Amazon Nova Act to control real browser sessions that navigate Yelp and AllTrails in parallel threads, then hit Google Maps for directions. The two searches run simultaneously and the results feed into a final Google Maps lookup for driving distance and time.
Challenges we ran into
The biggest issue was speed. Each Nova Act browser session was taking 4+ minutes because the agent had to navigate search boxes, apply filters, and browse through results step by step. We fixed this by pre-building search URLs to land directly on results and running the restaurant and trail searches in parallel threads, which cut the runtime roughly in half.
Accomplishments that we're proud of
Getting two browser sessions to run in parallel and finish independently before handing off to Google Maps actually cut runtime significantly. It also just works end to end, which took more debugging than expected.
What we learned
Nova Act is a browser automation tool, not an LLM. That distinction matters a lot when you're trying to extract structured information. Tight prompts with pre-built URLs make a bigger difference than any model setting.
What's next for Weekend Scout
Support for more activity types beyond hiking, better handling of bot detection, and a proper frontend so anyone can use it without running a script.
Log in or sign up for Devpost to join the conversation.