Hackathon ProjectStatus: Complete

Phone Finder (SpecShark)

SpecShark takes on decision overload when buying a phone. Instead of filters and a dozen review tabs, it asks at most five questions, chooses each one adaptively, and shows its work: the field of phones shrinks after every answer, and the eliminated phones get eaten by a shark.

My role
Hackathon project with a teammate; I rebuilt it afterward
Topics
Next.js · TypeScript · Recommendation · Web Audio · Hackathon

The Idea

Choosing a phone means comparing dozens of devices and still not knowing what to buy. The idea was an interrogator rather than a filter screen, and the key insight was that the elimination is the product: a recommendation that just prints an answer is hard to trust, while one that shows the field collapsing to a single phone is convincing, and fun to watch.

What I Built

Built at a 48-hour hackathon with a teammate. You pick an ecosystem, then answer questions on a scale. After each answer the remaining phones re-rank, the weakest are cut, and the shark eats them. The app explains why it asked each question and why it chose the winner, and a Surprise Me mode plays a whole run on its own for anyone watching.

How It Works

Each answer shifts a preference profile across 17 normalized phone attributes, and every phone is scored against it. The next question targets the attribute the remaining phones disagree on most, the question that splits the field best, and the engine stops early once the leader is clearly ahead. The 30 phones and 15-question bank live in plain data files, so either can grow without touching the engine. Every sound effect is synthesized live with the Web Audio API, with no audio files.

Challenges & Decisions

After the hackathon, I wrote down what the build was trying to do and where it fell short, then rebuilt it. Two bugs undercut the whole idea: the field count could go back up when the leader changed, and the shark sat on the one animation frame that renders nothing, so it never appeared. The rebuild fixed both and redesigned the visuals around the predator metaphor instead of a generic phone-app look.

What I Learned

A demo has one job, and the design should serve it. Writing out the original intent before rebuilding, and separating the good idea from the rushed execution, made the second version much more focused.

Current State

Complete and playable online.