Ashfaaq Farzaan

tie-hunt 🎯

🎯 Project Overview

tie-hunt is an ongoing research project exploring threat prediction using agentic hunting. It aims to forecast an adversary’s next steps by predicting MITRE ATT&CK® techniques that haven’t been found yet.

Built on the MITRE CTID Technique Inference Engine, this project currently adds trained models and a pure-numpy serving layer, laying the groundwork for an automated, agentic hunt pipeline.

🔬 Ongoing Research: Agentic Hunting

The core research focus is combining TIE’s predictions with autonomous agents:

  • TIE Ranking: The engine ranks candidate techniques based on observed behavior.
  • Agentic Validation: Local agents autonomously check each predicted technique against an EDR.
  • Feedback Loop: Confirmed techniques feed back into the model for the next predictive pass, making scalable lead testing possible.

Current work involves benchmarking this combination against TIE alone and agents alone.

🚀 Key Features

  • 🧠 Recommender System: Treats threat reports as users and techniques as items to suggest missing techniques based on co-occurrence patterns using WALS matrix factorization.
  • ⚡ Fast Inference: Features a pure-numpy serving layer (tie_serve) that imports in 84 ms and delivers sub-millisecond predictions, perfect for rapid agentic querying.
  • 🛠️ Ready-to-use Models: Includes pre-trained models and a CLI for inference.

🔗 Explore the Repository