Applied AI & Systems Engineer building production-grade AI systems at the intersection of edge computing, ML infrastructure, and autonomous systems. Completed an M.S. in Applied Artificial Intelligence at the University of San Diego.
Focus Areas: Edge AI Deployment · ML Infrastructure · Production AI Systems · Aerospace & Defense AI · Autonomous Systems · Cybersecurity AI
| Project | Description | Stack |
|---|---|---|
| PEREGRINE | Aerospace object detection system — CapsNet implementing Hinton 2017 dynamic routing, trained on 68,000 real satellite images, deployed on NVIDIA Jetson Orin with 2.65ms edge inference, autonomous investigation agent, live Prometheus/Grafana monitoring, and drift detection. Live at peregrine.bitshadow.dev | PyTorch, CapsNet, DIOR, ONNX, Jetson, FastAPI, Prometheus, Grafana, Docker, K8s |
| AXIOM | Defense-grade edge-cloud hybrid AI assistant — voice interaction, multi-agent debate, autonomous monitoring, holographic UI, 99.9% uptime since launch. Live at axiom.bitshadow.dev | Python, Flask, Docker, K8s, Jetson, Claude API, ElevenLabs, Prometheus |
| BitShadow | Live AI-powered phishing URL detector with ML classification and rule-based threat scoring | FastAPI, Streamlit, NLP, ML |
| Shadow Tribunal | Local LLM inference system running private AI models | Ollama, FastAPI, Python |
| Signal Sunday Bot | Production automation bot for messaging workflows | Python, Signal API |
| Adaptive Anomaly Monitor | Time-series anomaly detection — Isolation Forest, Rolling Z-Score, and LSTM Autoencoders | PyTorch, scikit-learn |
PEREGRINE — Capsule Networks for Aerospace Object Detection
Implemented Hinton's 2017 dynamic routing by agreement from scratch and benchmarked CapsNet against ResNet on 68,000 real aerial images from the DIOR dataset. Key finding: CapsNet outperforms ResNet across all 6 rotation angles tested with maximum advantage of +8% at 60° rotation — validating Hinton's thesis that capsule networks preserve spatial relationships that CNNs lose through max-pooling.
- CapsNet accuracy: 82.99% on full DIOR dataset
- Rotation invariance: CapsNet wins 6/6 angles tested
- Edge deployment: 2.65ms inference on NVIDIA Jetson Orin Nano
- Model size: 12.2MB ONNX — no cloud dependency
Core AI / ML — PyTorch · TensorFlow · scikit-learn · ONNX Runtime · CapsNet · LLMs · NLP · Computer Vision · Anomaly Detection · Speech Recognition
MLOps — MLflow · Prometheus · Grafana · Drift Detection · CI/CD · GitHub Actions · GitLab CI/CD · DevSecOps
Systems / Infra — Linux · Docker · Kubernetes (K3s) · REST APIs · FastAPI · Flask · Microservices · JWT · Rate Limiting
Edge / Embedded — NVIDIA Jetson Orin Nano · C++ · OpenCV · ARM Architecture · Multi-threading · NVIDIA Nsight Systems · CUDA · TensorRT · ONNX Runtime
Cloud / DevOps — AWS (Athena, EC2, S3, CloudWatch) · DigitalOcean · Cloudflare · CI/CD
Languages — Python · C/C++ · Bash · SQL · Java · JavaScript · Go (familiar)
- Original AI research — implemented Hinton 2017 CapsNet from scratch, validated against paper, benchmarked on real aerospace data
- Edge-cloud hybrid AI systems deployed on NVIDIA Jetson Orin Nano and production cloud simultaneously
- Full MLOps lifecycle — experiment tracking, model versioning, drift detection, auto-retraining triggers, audit logging
- Multi-agent AI architectures with specialized agents for autonomous investigation and decision making
- Production ML APIs with JWT auth, rate limiting, Prometheus monitoring, and auto-healing Kubernetes deployments
- Real-time voice AI pipelines — wake word detection, STT, LLM inference, ElevenLabs TTS under 2 seconds
- AI-powered cybersecurity systems and anomaly detection for infrastructure monitoring
Red Hat OpenShift & Kubernetes · AWS Cloud Foundations · Google Cybersecurity · Google Data Analytics · Cisco Network Automation
