Built and demoed DevDiff — a real-time pull request risk intelligence prototype.
In this video, I show how DevDiff analyzes PRs using:
20-rule static engine
Custom Random Forest ML scoring
Optional Groq LLM logic review
Live websocket findings stream
Developer-wise feedback learning (false positive → low priority → ignored)
What you’ll see:
Create/select project
Analyze GitHub PR
Real-time findings + risk score
History, scorecard, heatmap, developer profile
How past patterns and feedback reduce noise in future scans
This is designed as a full loop:
Detect → Score → Explain → Learn
If you want the source walkthrough + architecture breakdown, comment “ARCHITECTURE”.
