I document the systems I built, the research I tested, and the decisions that shaped the results. These studies cover cloud infrastructure, agent harnesses, applied AI, and automation, including what worked and what I would change.
Working systems, reproducible benchmarks, deployment history, costs, failure modes, and direct paths to source material.
Completed studies and source-backed project records, ordered for fast technical review.
Architecture, Supabase Auth, Stripe billing, Pinecone retrieval, Azure ML benchmarks, release diagrams, costs, failure modes, and remaining evidence gaps.
Stateful n8n migration, an OpenClaw personal agent with scoped environment control, private Azure networking, GitHub OIDC delivery, exact cost attribution, failure analysis, controlled teardown, and recoverable lower-cost redesign.
Benchmarks the ViperMesh Blender MCP and its agent harness against the original BlenderMCP baseline, with time, token, tool-use and visual evidence. Part One covers the Blender connection, not the wider Studio or neural 3D tooling.
Framework for agents to inspect real artifacts, trace failures, develop isolated harness candidates, validate regressions, and promote only evidence-backed improvements.
Handcrafted, community-curated, and synthetically generated n8n workflows for direct import, ML training, and retrieval-augmented workflow generation.
An autonomous purple-team IaC sandbox built for the AMD x lablab.ai hackathon, then expanded into an open-source DevSecOps alpha with reproducible attack, remediation, and evidence workflows.
Business and public platform used to deliver client automations, integrate AI/ML systems into workflows and projects, provide technical consultancy, and historically offer end-to-end OpenClaw setup.
Each future page should cover executive summary, problem, constraints, requirements, architecture, security, deployment, operations, cost, results, tradeoffs, failure modes, improvements, repo/demo links, interview explanation, and resume bullets.