🧵 How we built an AI detector that uses AI to detect AI-generated content

1/ The Challenge

With AI writing tools everywhere, verifying content authenticity is crucial. We needed a tool that could analyze text and identify AI patterns in real-time.

2/ The Architecture

We use Server-Sent Events (SSE) to stream results as they're generated. Text is chunked by sentences (max 1000 chars) and analyzed segment-by-segment.

3/ Pattern Recognition

Our AI looks for:

• Generic language ("it's worth noting", "delve into")

• Repetitive structures

• Unnatural phrasing patterns

• Overly formal or generic tone

4/ Confidence Scoring

Each segment gets a 0-100% confidence score. We calculate overall confidence by averaging AI-detected segments. This gives you granular insight into which parts are likely AI-generated.

5/ Rate Limiting

We use MongoDB atomic operations for rate limiting (3 requests/day per IP). This prevents race conditions and gracefully degrades if the database is slow.

6/ The Result

A free public tool that works for both logged-in and anonymous users. Real-time progress tracking, detailed segment analysis, and confidence scores.

Try it: sourcepilot.co/detector