Projects
Edge inference, agentic frameworks, and AI that runs in production for years, not demos. Alongside that, medical devices, clinical AI, cybersecurity, and technical consulting: systems built for environments where failure has consequences.
Open source
- SKIFF: Container Manager for Zero Trust Environments
- Open-source container manager built for GCP and zero-trust workstations. Avoids nested virtualization and privileged socket mounts. Talks to the daemon over SSH with nothing persistent on the container host. Four-tier test suite, 94% coverage floor, anti-pattern linter, and per-framework compliance evidence (OWASP ASVS, NIST SSDF, NIST CSF, SLSA, CIS Docker Benchmark). MIT-licensed.
- Agent Memory: Persistent KV Cache for Multi-Agent Inference
- MLX implementation of persisted 4-bit KV caches for multi-agent LLM inference on Apple Silicon, with background reloads and a demo app. Code for the Agent Memory Below the Prompt paper.
- Agentic Orchestration Framework
- Enterprise alternative to LangGraph. Lets non-ML engineers build production AI workflows. Open-source release in progress.
Production systems
- Scaling Agentic AI Across an Enterprise
- What it actually takes to move agentic technology from a pilot to organization-wide adoption: governance, security, enablement, and the cultural work nobody talks about.
- Production ML: Forecasting and Supply Chain Optimization
- Ensemble forecasting system designed for longevity, still running after a decade. Companion supply chain optimization algorithms for inventory risk-sizing, pricing integrity, and margin protection. Designed as integrated systems where ML, data engineering, and business logic are treated as a unified whole.
- Neural DSP for Edge Deployment
- Replacing traditional signal processing pipelines with learned models for embedded systems. Targeting Cortex-M class MCUs and small NPUs. Neurosymbolic approaches that preserve physics constraints while reducing compute by orders of magnitude.
- Stateful Agent Architecture for Low-Latency Retrieval
- Memory-safe context management for agentic workflows. Persistent state segmentation with retrieval from short contexts at low latency and low hallucination. Treats jailbreak detection as a buffer overflow problem.
- Classical Baseline vs. RL: Franka Pick-and-Place in Isaac Sim
- 150 lines of classical control logic for a Franka Panda doing randomized pick-and-place. No training, no reward function. The baseline tells you exactly where learning earns its complexity budget.
Safety-critical and clinical
- Agentic AI for Heart Failure Management
- Biomarker-driven heart failure system-of-care. Scalable to all patients in all care settings.
- Computer Vision & Audio Sensing for Implantable Devices
- Peer-reviewed publications in medical technology.
- IMU Dynamics Simulation with Bayesian Inference
- NumPyro-based simulation of noisy IMU measurements with known physics. SVI parameter estimation and smoothed calibrated state estimates for medical wearables.
Patents
- US Patent 7,936,274, Physical Identity Card Protection
- Commercialized and licensed.
- 14 Provisional Patent Applications, Soundness AI
- 14 provisional patent applications covering 21 invention families (2026), currently unassigned.
- 13 Additional Provisional Patents, Safety-Critical AI Architecture
- Coverage spans deterministic safety enforcement, cost-optimized inference routing, source attribution, regulatory-compliant update workflows, evidence-aware text processing, closed-loop clinical monitoring, multi-device AI coordination, population drift detection, and automated compliance verification. Patents cover edge deployment, clinical AI, and enterprise compliance. These are systems built for environments where failure has consequences.
For commercial products or licensing, contact me on LinkedIn.
Advisory
- Technical Expert Consulting
- IP litigation, licensing disputes, and technology valuation for manufacturers, investors, and legal teams. Nine years of translating engineering complexity into business decisions.
- Cybersecurity Advisory
- Identifying security and compliance risks from AI and data systems.
- Shadow AI Reduction Through Culture, Not Surveillance
- Organizations that frame AI governance as innovation enablement see less shadow AI than those that frame it as risk mitigation. Enablement-first governance frameworks.
Tools
- Software Employment Calculator
- An interactive model of how AI and layoffs change software-team output: new teams gain fast, established companies see muted impact.
