I’m a security engineer at Comcast, and most of what I do outside that is adversarial evaluation of LLM and agentic systems. Prompt injection that survives a real pipeline instead of a bare model, scoring harnesses that catch it, and what containment actually buys you once an agent can reach the network.
That work lands upstream where it can. Prompt injection and credential-leak scoring in PyRIT, a Best-of-N jailbreak implementation, a PDF injection probe for garak, crash and hardening fixes in Google’s OSV tooling. I’d rather put a technique into a framework other people already run than publish a tool nobody installs.
The rest of my time goes to building. sentinel-maas is an endpoint agent I wrote from scratch in Go, with an adversarial testing module that fires synthetic ATT&CK techniques through the real detection path to prove the pipeline works. Sleuthgraph is an open-source OSINT investigation workbench, graph-native and built around pivots. I lead ThinkEngine, which is being open sourced at Comcast.
CTFs regularly, mostly with my team at Comcast, plus OSINT search parties on my own. Bug bounty on Bugcrowd. Currently working toward OSAI, OffSec’s AI red team cert. Attacking LLMs, RAG pipelines and multi-agent systems is the same problem I spend the rest of my time on, so it lines up.
MSE in Software Systems and Cybersecurity from Penn. CISSP, CEH, Security+, CCNA. One preprint so far, on out-of-bounds memory access in Zig across build modes.
This site is the working notes. Research is what I’m on now, projects is what I’ve shipped, notes are short findings: what I tried, what held, what didn’t, and what I’m not claiming.
Public work only. Nothing from the day job, nothing under review.