US Congress Scorecard
Overview
My role: System redesign to deployed system, run and maintain.
Challenge: Make a vibe-coded scorecard about US Members of Congress deployable
Solution: Complete rewrite of the Python spaghetti code into a structured, secure system
Technologies: Quarkus / Java 21, React / TS
Vibe
The codebase received was the result of early ChatGPT 3/4 or Sonnet 3.7 generations. It had original structural patterns that corresponded to a well-designed system. As features were added, each run deposited layers upon layers of new code into various places across the codebase. The old code was left behind, sometimes still active, sometimes dead, sometimes half-dead: endpoint still available, but nobody to call it.
Later changes showed the fingerprints of Sonnet 4.5: Lots of unmistakable green ticks with triumphalistic reports about new features that may or may not have been implemented.
Testament to the limitations of AI assistants, even with the best of intentions.
Structure
The target deployment environment was a WordPress site needing both administrative and public access to the scorecard. As making the original codebase deployable meant splitting functions, ensuring security, removing dead code and also help it live in a WordPress environment, whilst at the same time trying to fix bugs, too, it was simpler to start afresh and re-implement the features rather than to keep pulling out one spaghetti at a time.
The new version is, of course, built using agentic AI. But with clear architectural decisions, strong test coverage, alignment to the WordPress requirements from the get-go, this one remains a clean codebase.
The easy platform for running projects of any size.
To discuss tailored projects, please connect on LinkedIn.