How a developer used Claude Code to turn an internet trend into a working web app at our eight-hour workshop with Anthropic
The Event: A Front Row Seat To Innovation
Aura is the internet’s unofficial measure of coolness. TikTok built an ecosystem around it: creators rating each other’s presence, comment sections alive with «+100 aura» and «aura drain,» a generation of users who instinctively understand the mechanic.
On August 29, Wizeline and Anthropic brought a room of engineers together in Guadalajara for an eight-hour Claude Code Workshop. Each team had free rein to build whatever they wanted. Most stayed in practical territory. Then Luis Gil, a developer from HCLTech, walked to the front and demoed Aura Farmer.
The Demo: Aura Farmer
Aura Farmer is a browser-based web app that converts physical poses in front of a camera into points, combos, and social-ready video clips. The demo landed immediately. The concept is simple; the engineering underneath it is not.
The app tracks 12 distinct movements, each with its own scoring logic, cooldowns, combo multipliers, and dynamic rank thresholds. Four game modes shape how you play:
- Farm Mode. Timed runs of 15, 25, or 40 seconds
- 1v1 Duel. Two players with separate tracking and scores on the same screen
- Tutorial. A 12-step guided practice session
- Free Mode. No timer; play until you stop
All pose estimation runs locally on the device via MediaPipe. Video stays with the user until they decide to share it. The final output is a 9:16 vertical canvas that composites the camera feed, the heads-up display, and visual effects into a single clip, recorded exactly as it appeared during play.
The Architecture: Designed For Modularity
Luis built Aura Farmer on React 19, TypeScript, and Vite 7, with MediaPipe Vision handling pose estimation. The stack is not unusual. What made the app work was the engineering discipline underneath.
Three decisions kept it stable outside demo conditions. The movement detectors use body-proportion geometry rather than pixel positions, so the app performs consistently regardless of how far the player stands from the camera. When multiple poses register in the same frame, only the highest-confidence one scores, which eliminates double-fires and incoherent results. In 1v1 mode, player identity is assigned to each body before the smoothing filter runs, so two players can cross paths without the system losing track of who is who.
The recorded clip captures the session exactly as it appeared: camera, HUD, particles, and score, composited at the moment of recording. Because all processing runs on-device, privacy is built in by default.
The Developer Workflow: The Human In The Loop
The most transferable part of Luis’s presentation was the process.
He did not hand Claude Code a product brief and wait for a finished app. The workflow ran in short, deliberate cycles. Each round started with a precise intention: detect this movement, score it, add a visual effect. Claude Code read the codebase, identified the affected modules, wrote the TypeScript, and ran typechecks and build verification. Luis then stepped back in: physical testing of the movement, evaluation of the feel, correction of false positives, and refinement of the UX before the next cycle began.
The AI handled execution. The developer handled judgment. Claude Code compressed the cost of each iteration; Luis spent that time deciding what the app should actually do and how it should feel to play.
Next Steps
Watching Aura Farmer in action proved that AI coding tools reduce the heavy cost of iterating on architecture. The real advantage is that developers gain the freedom to focus entirely on product direction and user experience. To learn how to apply these workflows to your own engineering teams, join our next Wizeline technical workshop.
