ljon

What it is: Questr is a static, zero-backend web application that turns any location on Earth into an interactive, gamified travel experience. Users drop a pin anywhere to immediately fetch real-time geographic metadata (weather, elevation, sun times), play hyper-localized landmark trivia, and build an AI-orchestrated travel journal. The Problem It Solves: Friction-Heavy Travel Logging: Traditional travel tracking requires heavy databases, slow mobile app setups, and invasive tracking permissions. Questr delivers zero-friction exploration entirely on the static edge. Brittle AI Content Generation: Standard multi-modal workflows frequently produce poorly styled layouts, mismatched themes, or text clipping when building dynamic digital scrapbooks. The "Prompt Engineering Collapse" in UI: In most LLM pipelines, if a model misjudges a design or visual theme, the user gets an ugly, broken asset. How We Solved It (The Cursor/Agentic Twist): We built an autonomous Optimizer-Evaluator loop using OpenRouter vision models and a Python/Pillow rendering engine. When a user drops 3–6 trip photos, our factory pipeline spins up. The Evaluator model sets a strict layout, color palette, and thematic success metric. The Optimizer generates candidate rendering permutations, which the Evaluator critiques and scores against the criteria until a flawless, beautifully balanced travel-journal collage is generated. We also engineered a Git-driven path: users can commit a template markdown file to GitHub, and a GitHub Actions workflow automatically emulates the reference layout to generate the journal entry completely offline. The entire front end runs instantly on Cloudflare Pages, pulling zero-dependency open data, while the secure agentic engine runs smoothly in a lightweight, self-hosted container.