ActivityScout

Find programs kids love. Parents can afford. Applications made easy.

Comment

Inspiration

Finding extracurricular activities for my younger sibling, and understanding how difficult it is for many families across Toronto and North America.

What it does

ActivityScout uses Claude AI to help families find kids' programs they can actually afford. It solves three problems: (1) Finding hidden financial aid - AI searches program websites for scholarships and subsidies that traditional search engines miss, (2) Smart matching - analyzes what programs value (teamwork, creativity, diversity, etc.) and matches them to each child's personality and strengths, (3) Easy applications - extracts direct registration links so parents don't waste 30 minutes hunting through websites. The AI performs 3-4 web searches per query to find opportunities families would never discover on their own.

How we built it

  • Frontend: Next.js 16 with TypeScript, Framer Motion for animations, and WCAG AA accessibility compliance
  • AI: Claude Sonnet 4.5 with the web search tool - this lets Claude autonomously search the web, visit program websites, and extract detailed information
  • Smart features: 4-factor match scoring algorithm (interests, strengths, needs, budget), 9-dimensional program values analysis, Server-Sent Events for real-time streaming
  • Enhanced data: Google Places API for ratings and reviews, geocoding for distance calculations

Challenges we ran into

  • Rate limits: Claude's web search generates long prompts - hit the 10,000 tokens/minute limit during testing, had to implement caching
  • Parsing Claude responses: Claude sometimes returned extra text before the JSON - had to write robust regex extraction and trailing comma cleanup
  • Web search effectiveness: Initial prompts didn't find financial aid info - solved by explicitly instructing Claude to search for keywords like "scholarship," "sliding scale," "subsidy"
  • Accessibility: Initial design had glowing hover effects and poor colour contrast - completely reworked to support motion preferences and
    WCAG AA standards

Accomplishments that we're proud of

  • Multi-search strategy: Unlike competitors that do one search, we perform 3-4 searches with varied terms to find hidden opportunities
  • Financial aid detection: First tool to explicitly surface affordability information that's buried on program websites
  • 9-dimensional values analysis: AI analyzes what programs truly value (diversity, teamwork, leadership, etc.) by reading their mission
    statements and content
  • Real human impact: Solves a real problem - many parents who want to enroll kids in programs can't due to cost and accessibility barriers

What we learned

  • Claude's web search tool is powerful: It can autonomously visit websites, extract structured data, and analyze content - way beyond
    simple search results
  • Streaming UX matters: Real-time updates ("Performing web search 2...") make long AI operations feel responsive and build trust
  • Prompt engineering is critical: Spent hours refining the prompt to get consistent JSON output with all required fields
  • Accessibility isn't optional: Motion preferences and keyboard navigation matter - disabled users deserve great experiences too

What's next for ActivityScout

  • Multi-language support: Spanish, French, Mandarin, Arabic translations to reach immigrant families who need this most
  • SMS notifications: Application deadline reminders and new program alerts based on saved profiles
  • Partner integration: Direct API connections with recreation departments and program providers for one-click signup (not just links)

Built With

Share this project:

Updates