2026Web AppSHIPPED

Trend Intelligence

Full-stack AI trend intelligence platform for Central America and the Caribbean. Two verticals — food & beverage and pet nutrition — from a single codebase. Type a category and get a complete strategic brief in under 3 minutes, grounded in live signals from Google Trends, YouTube, Reddit, and Firecrawl.

What it does

  • Two dedicated search experiences — food & beverage and pet nutrition — from a single codebase, each with its own brand-level output section
  • Four data sources run in parallel before Claude writes a single word: Firecrawl (3 simultaneous searches across regional market data, global benchmarks, and industry reports), Google Trends (12-month search interest across 9 countries with direction signals and rising related queries), YouTube (top Spanish-language videos from the last 6 months ranked by views), and Reddit (real consumer conversations surfaced via Google Custom Search)
  • Each query produces a complete strategic brief in under 3 minutes: three classified trends (Alto Momentum / En Ascenso / Emergente) with regional insight and a global benchmark for each, emerging signals with cross-regional context, three key stats plus a consumer quote, strategic recommendations, consumption frequency, use occasions, and three to five recipes adapted to the Central American palate with brand activation tips
  • Country chips color-coded — green for strong signal, amber for moderate, grey for minimal — with a smart search-term hint when low scores are terminology-driven
  • One-click print and PDF export with no dark backgrounds, broken layouts, or unreadable text — section headings stay with their content, major sections start on fresh pages
  • Full search history stored per vertical so returning users resume where they left off
  • "Copiar resumen" button generates clean plain-text output ready for presentations and briefs

Why it matters

  • Reduces a full trend brief from 2–3 hours of desk research to under 3 minutes — regional search data, video trends, consumer conversations, and web signals synthesized in one pass
  • Four data sources in parallel add no perceived load time — all four calls run concurrently; the user waits for the slowest source, not the sum
  • Grounded in Central America and the Caribbean specifically — Google Trends scores are interpreted relative to the region's naturally lower search volumes, not US or European baselines
  • Graceful degradation: if any data source is unavailable, the analysis continues without it and the brief is still complete — no errors, no empty pages
  • Source attribution on every trend, stat, and recipe so teams can trace claims back to their origin

Architecture

  • Vite + React 19 single-page app with two route contexts — one per vertical — sharing a common analysis pipeline
  • Vercel serverless functions handle each data source independently; an orchestration layer assembles the four result sets before calling Claude
  • Claude Sonnet with `tool_use` for deterministic structured JSON output across all brief sections — no regex parsing of prose
  • Google Trends API queried for 9 countries simultaneously with direction signal computation (rising / steady / declining based on slope over the period)
  • YouTube Data API fetches Spanish-language content from the last 6 months, sorted by viewCount, with channel attribution
  • Reddit surfaced via Google Custom Search scoped to reddit.com — real consumer language without requiring a Reddit API key
  • Firecrawl runs 3 parallel searches per query: regional market signals, global benchmarks, and category-specific industry reports
  • Print stylesheet strips interactive elements, forces black text on white backgrounds, and injects page breaks before major sections

Philosophy

  • Real signals from four independent sources, cross-validated by Claude — not one curated database with a chat interface on top
  • Parallel execution as a design constraint: adding three new data layers added zero seconds to perceived load time
  • Every claim in the brief is traceable — source attribution is structural, not a footnote
  • React 19
  • Vite
  • Tailwind CSS
  • Claude Sonnet
  • Firecrawl
  • Google Trends
  • YouTube API