Building toward production-grade AI systems at the intersection of LLMs, retrieval, and security. U.S. Air Force veteran turned engineer, currently a Security & Awareness Analyst at Centene.
I'm building toward a career at the frontier of AI engineering — focused on LLM application development, retrieval-augmented generation (RAG), and the MLOps infrastructure that takes a system from demo to production. I build with the OpenAI and Anthropic APIs and care about what it takes to actually ship: latency, observability, safety, and architecture that holds up under real load.
My foundation is unusual on purpose: U.S. Air Force Security Forces veteran, B.S. in Computer Science, M.S. in MIS with an Information Security specialization (4.0 GPA), AWS Cloud Practitioner, Google Cybersecurity certified, and hands-on experience in Python, SIEM, and threat detection. That security-first instinct shapes every system I ship.
Latency budgets, observability, graceful degradation. AI systems that survive contact with real users.
Threat-modeling, prompt injection, adversarial ML, zero-trust thinking baked in from day one.
On-device inference. Zero cloud dependency when the workload doesn't need it. Your data stays with you.
Five years in Air Force Security Forces — mission-critical execution under pressure. It carried over.
Twenty years of building, defending, and now teaching machines to do the same. Every role compounds — military discipline, fleet-scale data systems, enterprise security, and now applied AI.
A live look at what's running on my own hardware right now — production-grade edge AI, distributed inference, and the AI/security research backing it. A polished portfolio site goes here next.
Production-grade surveillance running entirely on local hardware. Reolink PoE camera streams into a Raspberry Pi 5 + Hailo-8 NPU executing YOLOv8s at 25fps with ~100ms inference. GStreamer pipeline, SQLite event logging, Flask REST API (9 endpoints), Telegram alerts, and systemd services.
Edge system over LAN. Pi #1 handles inference and serves a detection API. Pi #2 runs Ollama (Llama 3.2) and a SOC-themed dashboard consuming that API. Fully on-device — zero cloud dependency, zero exfiltration.
Security Operations Center-style monitoring console with a dark theme, real-time KPI tiles, Plotly visualizations, modal AI assistant, and forensic history view. Streamlit MVP shipped; React/JSX migration in flight for production polish.
Local Llama 3.2 wired to a context-injection pipeline that feeds live detection stats — enabling natural-language queries like "How active has the dog been?" entirely on-device.
Active development against the OpenAI and Anthropic APIs — structured outputs, prompt engineering, agentic workflows, and the production patterns that take a prototype to a system you can trust.
Multi-agent systems built with the OpenAI Agents SDK, CrewAI, LangGraph, and AutoGen. Includes a Deep Research agent and a 4-agent engineering team coordinating real work.
Vector database integration (Chroma, FAISS) for context-aware AI applications — chunking, embedding strategy, retrieval evaluation, and the observability layer to keep it honest.
React/JSX work from GRIND (a WHOOP OAuth2 wellness PWA) and The Grail CLT (AI sneaker valuation) — applied to AI-native dashboards. Real product surfaces, real auth flows, real data — not toy demos.
Cybersecurity lens applied to AI: prompt injection, adversarial vision attacks, secure LLM deployment patterns. Backed by 5+ years of threat analysis experience.
Detailed case studies — architecture diagrams, evaluation results, and the stories behind the systems — are landing here.
Open to AI engineering roles, contract work, and serious collaborations on RAG, edge inference, and secure LLM systems.