Outside of my day job, I independently build production-grade AI systems. It started as curiosity about what these tools could actually do. It turned into a serious applied research practice spanning the full ML lifecycle — from model fine-tuning to system architecture to bare-metal infrastructure.
This is not theoretical anymore; I build things that run.
Autonomous agent systems
Built autonomous agent systems with specialized agents for different business functions. Custom LLM fine-tuning workflows for open-source models (Llama), training data pipeline development, and orchestration frameworks that coordinate multiple AI agents on complex tasks.
Zero-trust AI security
Designed and prototyped an identity management system using information silo architecture, opaque identifiers, and field-level encryption. Exploring security-first approaches to AI platform governance — how to give AI systems access to what they need without exposing what they should not see.
LLM lifecycle tooling
Developed production-grade tooling for the full LLM lifecycle: prompt optimization frameworks, ensemble routing across multiple providers (OpenAI, Anthropic, open-source), real-time cost tracking, and RAG (Retrieval-Augmented Generation) vector engine implementation.
AI product infrastructure
Prototyped a software licensing platform with multi-tenant architecture and usage-based billing, exploring how AI products get distributed, metered, and monetized at scale.
Business process intelligence
Designed and deployed intelligent automation pipelines using n8n for business process orchestration, from raw data ingestion through analysis to actionable output. The focus is on turning unstructured operational noise into structured intelligence without requiring human review at every step.
Self-managed infrastructure
Built and manage a personal lab environment: 10G networking (MikroTik), TrueNAS storage clusters, GPU compute servers, Raspberry Pi edge clusters, and Kubernetes orchestration. All research runs on infrastructure I own and operate.
In the open
Some of the research ships as public work: The ArcGIS Compendium (a free eight-volume plain-English reference), the n8n Compendium and Automation Decision Guide, and the build logs in Field Notes. The rest shows up as products on the Portfolio page.