We build, train, and defend
AI systems. Including yours.
We help security teams defend agentic systems and LLM integrations, build custom models and data pipelines, and keep those controls in place as the stack changes. Three things, done thoroughly: protect what you have, build what you need, and make it stronger over time.
Leadership
Zachary Hunsaker
BEAM Architect
zachary@sassafras.aiZac holds a Master's in Applied Intelligence from Georgetown University, leveraging open-source intelligence to strengthen security posture and inform strategic decisions.
His career spans incident response and security architecture, including threat modeling for the CloudFormation team at AWS, hardening Infrastructure as Code at scale.
He's also the author of over a dozen open-source Elixir libraries for the security community, including ExEcrime, Intel471Ex, domaintwistex, BraveEx, and builtwith — tools designed to make intelligence gathering and analysis faster and more accessible. Sassafras AI is the continuation of that work: bringing the same rigor and craft to defending agentic systems, hardening LLM integrations, and training models that hold up in production.
Philip Taggart
Vanguard Engineer
As a Senior Manager of Cyber Threat Engineering, Philip built an LLM-powered intelligence platform on AWS that automates intelligence extraction, maps actor-victim-industry relationships in Neo4j, and generates daily executive briefings — from priority intelligence requirements to country-specific threat assessments. He also deployed a TensorFlow neural network for automated domain risk scoring in brand impersonation detection.
His background spans defensive cyber operations at the JFHQ-DODIN Intelligence Directorate, integrating across 40+ federal agencies, and over a decade of operational leadership as an F/A-18 mission commander with 40+ combat missions. He brings that same operational rigor to engineering: systems that work under pressure, at scale, when it matters.
Spencer Cutrell
Security Philosopher
Spencer has formal experience and training in logic, low level computer systems, philosophy among other things. He is fantastic at finding solutions to complex problems. Those skills have been utilized in critical industries like transportation and financial services to protect both life and property. Spencer is particulary situated in the development of ethical AI programs or fine tuning agentic workflows to optimize for logical reasoning.
He holds a philosopher's instinct for first principles and an engineer's instinct for shipping — and at Sassafras AI, those two instincts are the same one.
What We Do
Defend Agentic & LLM Systems
Agents and LLM integrations need production-grade controls. We help you shore up prompt handling, tool permissions, data paths, and autonomy — then put monitoring and guardrails in place that actually hold.
Build Custom Models & Neural Networks
From data collection and labeling to architecture design, training, evaluation, and deployment. We build classifiers, anomaly detectors, and sequence models that work in production — not just in benchmarks.
Data Engineering & Labeling
Good models need good data. We build ingestion pipelines, annotation workflows, and feature stores — then structure, label, and version the datasets your models will train on.
AI-Augmented Security Tooling
Intelligent alert triage, automated IOC enrichment, correlation engines, and analyst copilots. We build the tools that let your team work faster and with more confidence, powered by the same AI we help you defend.
Traditional ML & Data Science
Not everything needs a neural network. We use the full spectrum — regression, clustering, decision trees, time-series analysis — to solve the problem with the right tool, not the trendiest one.
How We Work
Learn your landscape
Your stack, your data, your threat model, and where your team spends time. We don't start with solutions — we start with understanding.
Design, build, and deploy
Custom models, hardened agentic workflows, data pipelines, or security tooling — whatever the problem requires. We build it, verify the controls, and put it into production.
Operate, monitor, and evolve
Models drift. Threats shift. Agents need their guardrails kept current. We stay engaged — monitoring performance, retraining when needed, and adapting the defenses as your environment changes.
Open Source
We maintain over a dozen open-source Elixir libraries for the security and intelligence community — OSINT collection, threat intel APIs, domain analysis, and more. The same philosophy drives Sassafras AI: build tools that make the work faster, more rigorous, and more accessible.
Ready to shore up what's next?
Custom models. Hardened agents. Real data pipelines.
Zachary Hunsaker | Sassafras AI
zachary@sassafras.ai