HeimdallQC
An automatic AI and plagiarism detection service
Overview
A content quality control service which automatically detects AI and plagiarism use in a user's website. Designed to prevent legal issues and SEO punishment over undisclosed AI use/duplicate content. Acts as a kind of digital circuit breaker, the last line of defence before issues arise. Uses Winston AI for the AI + Plagiarism detection.
Tech Stack
TypeScript
Next.js
TailwindCSS
Python
FastAPI
PostgreSQL
Key Decisions
1.
Main Algorithm
In the early stages, I experimented with flagging suspicious content via monitoring user behaviour. While this could have worked in theory, it wound up being too difficult to implement properly and felt too intrusive. I opted for a safer approach, more in line with AI/plagiarism prevention rather than surveillance.
2.
Removing Suspicious Text
Originally planned to modify suspicious text using either an 'AI Rewrite' or 'Auto-citation'. These generated new content for the detected texts, though they posed more issues than they solved. Landed on just implementing an automatic removal for safety.
3.
Feature Plans
Decided to add plans instead of just one feature set to expand utility and appeal to more users. The 'Extrinsic' plan was for detecting suspicious content being submitted to the user's website, while the 'Intrinsic' plan automatically audits their site on a schedule.