Agent workflows that stay reviewable
Queue-driven coding agents, GitHub App wiring, review loops, and debugging flows that keep autonomous work tied to real engineering checkpoints.
Austin, TX | AI tools, agent systems, product engineering
I build AI tools people can actually use.
I am a full-stack engineer who likes the unglamorous parts of AI work: memory that can be inspected, agents that can recover, interfaces that operators understand, and enough testing to know when the magic is faking it.
Selected projects
Recent work that made AI systems easier to run
Reliability tests for coding agents
Benchmarks for coding-agent runtimes, tool access, sandbox behavior, failure recovery, and the operational details that decide whether agents are useful.
Prompt-guided image workflows
ComfyUI and RAW photo workflows for prompt-guided editing, image cleanup, and controlled visual iteration without losing the source material.
Open-source AI
Open-source memory research and agent infrastructure
A lot of my AI work starts with one practical question: can I make this agent remember better without making it harder to trust? That has pushed me toward local memory systems, MLX experiments, graph cleanup, benchmarking, and tools that show their work instead of hiding it.
ClawFactory
A local runtime for OpenClaw bots with visible state, logs, secret handling, encrypted snapshots, controller controls, and emergency recovery. It is the kind of infrastructure I want around agents before I let them touch anything important.
OpenClaw delta-mem MLX plugin
An OpenClaw plugin that connects agents to Apple Silicon MLX memory experiments and a trained Hugging Face adapter. The goal is simple: make agent memory more testable, faster to iterate on locally, and honest enough to benchmark.
Spellcheck
A ClawHub skill that catches obvious spelling, speech-to-text, and project-vocabulary mistakes before an agent acts on them. Cleaner names and prompts help graph embedding normalization, reduce noisy memory edges, and improve LLM accuracy.
Skills
The stack I use when the AI work has to become real software
LLM orchestration
RAG / retrieval
Agentic memory optimization
Memory benchmarking
Graph embedding normalization
MLX
Apple Silicon inference
Hugging Face adapters
Rust
Vector search
Tool calling
React
Next.js
TypeScript
Python
Flask
Cloudflare Workers
OpenTelemetry
Playwright
Cypress
Docker
GitHub Apps
ComfyUI
Experience
Enterprise product engineering, now focused on AI tools
AI Implementation Consultant / AI Tooling Engineer
Building practical LLM and agent workflows for planning, coding, review, research, and operations, including ClawFactory as a local control plane around OpenClaw.
Lead Software Engineer / Front-End & Full-Stack Engineer
Built VMware at IBM Cloud product surfaces across React/Redux, reusable design-system components, Cypress testing, accessibility work, and platform/security maintenance.
Full-Stack Product Engineer / UI Architect
Owned Infinite IO's admin console, dashboards, operator workflows, frontend architecture, real-time diagnostics, and supporting Python/Flask services.
Frontend Web Developer
Built healthcare analytics widgets at The Advisory Board Company / CRIMSON and maintained high-traffic education product surfaces at MyEdu.
Archive
The old handmade site is still here
I kept the 2013 version around as an artifact. It is not the current pitch, but it is part of the trail.