Systems Security Engineer & AI Agent Defense Architect Bridging low-level systems engineering (kernel tracing, binary reversing, memory safety) with deterministic security guardrails for autonomous AI agents (Model Context Protocol, AST verification, and fail-closed sidecars).
I contribute verified bug fixes, kernel tracing updates, memory safety patches, and regression suites to foundational open-source security and reversing tools:
angr/angr (Binary Analysis & Symbolic Execution Platform)
- PR #7353: Resolved unhandled
IndexError: bytearray index out of rangeinUltraPageduring symbolic memory state merging across 4096-byte page boundaries (Issue #6314, superseding stale PR #6315). Enforced candidate offset clamping in_ultra_changed_candidates, aligned concrete chunk stepping, and hardened_containswith defensive out-of-bounds guards; authored comprehensive multi-page regression coverage intests/storage/test_memory_merge.py.
iovisor/bcc (BPF Compiler Collection)
- PR #5546: Migrated
tools/killsnoopfrom brittlesys_killkprobes to stable kernel tracepoints (sys_enter_kill,tkill,tgkill). Resolved breakage introduced in Linux 5.11+ (Issue #3592) and brought the Python implementation into parity with modern libbpf-tools.
- PR #27060: Resolved Heap Out-of-Bounds read (CWE-125) and buffer desynchronization in
rabin2ELF section resize (-O r) and permission patching (-O p), enforcing fail-closed parameter validation and cross-endian safety (Issue #15449). - PR #26852: Resolved SIGSEGV memory corruption in the PDB parser for enums with modified underlying types (
eLF_MODIFIERpointer cast fix, Issue #12924). - PR #26862 & testbins #146: Built and upstreamed the minimal binary PDB test fixture and unit tests for type parsing.
- PR #26771 (commit
704e2e8), PR #26617 & PR #26606: Multi-stage refactor of math expression error handling acrosscmd_seekandcmd_writecore commands (Issue #23362), eliminating silent invalid writes and adding unit test coverage.
A context-aware security sidecar protecting autonomous AI agents against logic and injection attacks.
- stdio MCP Security Proxy: Intercepts and validates Model Context Protocol tool invocations using per-session sliding-window tracking.
- Hybrid Defense: Pairs local deterministic "Iron Dome" rules with LLM intelligence under strict fail-closed security invariants.
- Built with concurrency-safe rate limiters, Trivy CI container scanning, and structured telemetry.
Autonomous DFIR framework for SIFT Workstations orchestrating memory forensics.
- Orchestrates Volatility 3 and FLS via a multi-agent Finite State Machine (FSM).
- Uses a deterministic sieve and "Hard Grounding" citation layer to eliminate LLM hallucinations, generating traceable MITRE ATT&CK incident reports.
Agnostic security verification platform and repository profiler.
- Performs automated architecture discovery and dynamic AST fallback sandboxing.
- Audits codebases against NIST AI RMF, OWASP LLM Top 10, and MITRE, emitting compliant SARIF 2.1.0 reports.
Autonomous self-optimization daemon and dynamic MCP tool host.
- Profiles, mutates, and tests routines inside isolated RAM-disk sandboxes using AST mutators and dual-phase scheduling.
- Copilot Enterprise Privilege Escalation: Technical report detailing container path hijacking vulnerabilities leading to root escalation in Microsoft Copilot Enterprise environments.
- Agent-Native Threat Architecture: Threat models for context poisoning and agent hijacking mapped against NIST and ISO/IEC frameworks.
- Languages & Systems: C, Python, Rust, Bash, Linux Kernel & Tracing (eBPF/BCC, tracepoints, kprobes), QEMU/KVM
- Binary & Reverse Engineering: angr (Symbolic Execution, UltraPage Memory), radare2, Volatility 3, GDB, PDB internals, Memory Layout Analysis
- AI Security & Protocols: Model Context Protocol (MCP), LLM Guardrails, AST Parsing, SARIF, NIST AI RMF, OWASP LLM Top 10, MITRE ATT&CK