FaultFindr

CATEGORIES

Go Deeper

FaultFindr is built around a simple idea: every strength has an inevitable downside. I designed a full 34-theme mapping system that reframes each trait as its cynical counterpart, with consistent logic and tone across all outputs. The system has two layers. First is a deterministic mapping layer that converts a user’s strengths profile into a structured “fault profile,” including naming, definitions, and prompt seeds. Second is an LLM generation layer that turns that structure into a personalized, roast-style report that feels specific rather than templated. The interesting part is the balance between control and emergence. The mapping ensures coherence, repeatability, and a clear point of view, while the model adds variability, voice, and psychological realism. That combination makes the output feel both authored and surprisingly personal. It is intentionally positioned as satire, but it functions as a behavioral mirror. Users tend to recognize themselves in the output, which creates a mix of humor and insight. From a build standpoint, it is a lightweight, AI-native web app deployed via Replit, with the focus on prompt design, taxonomy design, and output quality rather than heavy infrastructure.

Stack Used

Replit (full-stack hosting), Node.js/Express backend, React front end, OpenAI GPT and Anthropic Claude models for generation, plus a custom deterministic mapping layer for strengths-to-fault logic.