SIFT-BENCH: Autonomous IR Accuracy Benchmarking System

AI-powered IR agent that analyzes forensic evidence autonomously, measures its own accuracy against ground truth, and self-corrects — achieving 100% accuracy with 0% hallucinations.

Comment

posted an update —

Major Update — Version 2.0

Just pushed significant improvements to SIFT-BENCH:

  • Added CASE-002: Ransomware Attack test case
  • Both cases now scoring 100% accuracy with 0% hallucinations
  • Overall average: 100% accuracy across 2 different incident types
  • Improved agent prompts for better keyword coverage
  • Updated architecture diagram and documentation

Results summary:

  • CASE-001 (Malware Infection): 100% accuracy | 0% hallucinations | EXCELLENT
  • CASE-002 (Ransomware Attack): 100% accuracy | 0% hallucinations | EXCELLENT
  • Overall: 100% accuracy | 0% hallucinations

GitHub: https://github.com/Farhanahmadansari0173/sift-bench

Log in or sign up for Devpost to join the conversation.