Projects
Hugo RL-LLM
Production-grade foundation platform for LLM-enhanced Reinforcement Learning with reproducibility, checkpointing, benchmarking, LLM frozen mode, and a plugin system.
Strike
AI-augmented security testing framework in Rust: agent-based vulns discovery/validation, checkpointed workflow engine, LLM routing, CI policy gates (SARIF 2.1.0), and orchestration of nmap/nuclei/sqlmap/etc.
Trace
Hybrid autonomous SWE-Bench agent that reduces context overflow and file retrieval failures using AST-aware search and iterative refinement.
Adversaria
Rust-based adversarial testing harness for LLMs with multi-provider support and structured risk scoring.
Aegis
Deterministic capability firewall for LLMs using taint tracking and policy DSL enforcement instead of prompt filtering.
Zenta
Multi-language CLI and SDK suite for scanning and migrating classical crypto to NIST post-quantum algorithms in ML infrastructure.
ETH.id
Zero-knowledge document verification system combining local privacy filters with LLM semantic verification.
GrabKit
High-quality, resilient web content extraction toolkit that converts messy pages into clean, main-content outputs for AI systems using domain templates, layered heuristics, a proven parser, and optional browser rendering.