Python workflow
I use UV for all Python projects because it is fast, lightweight, and keeps environments simple. It fits AI-assisted development well because setup is quick, dependency handling is clean, and project bootstrapping stays low friction.
James Clark · DevBuddy
I build real systems with AI for real-world use.
I build with AI, then run the systems, test them in the field, and make sure they hold up where they actually matter.
About
Manager, operator, builder
I build tools for real environments, not just demos. My work is shaped by hands-on experience with retail POS systems, support operations, field troubleshooting, and the workflow friction that shows up when systems meet actual users.
I use AI as part of the development process, but the point is never to generate code for the sake of it. The point is to build something useful, pressure-test it in the real world, and make it hold up where it actually matters. My background is rooted in the Retail Data Systems environment, where speed, clarity, and reliability matter more than theory.
Tools
Tools of choice
Built for speed, low friction, and real-world iteration.
Frontend + experience
I am not locked into one language or framework. I choose the stack that fits the job, especially when the interface needs to feel polished, cinematic, and intentional instead of looking like another flat internal tool.
Core stack
PythonUV, FastAPI, automation workflows
JavaScriptThree.js, interactive UI direction
Graph LayerNetworkX, connected operational context
FrontendHTML, CSS, cinematic interface polish
Workspace AutomationGoogle Apps Script, Sheets, Chat workflows
System ToolingGo for monitors and support-side utilities
Projects
Selected work
A mix of public proof and private operational systems built for actual use.
RDSCore – Cinematic Command Center
A search-first operational command center designed to replace scattered CRM views, ticket history, vendor references, employee location sheets, and support context with one unified system. Built to feel clean, cinematic, and deliberate, while bringing critical information forward faster without unnecessary friction.
After Hours Emergency Alert System
One of the first tools I built with AI. I run a helpdesk where after-hours calls route to the on-call tech, but voicemail alert emails were often handled without anyone replying "got it" for the team to see. I built a Google Apps Script workflow that watches for those conditions and alerts a shared Google Chat room so the team knows a call may still need attention. It gave us a practical way to catch misses, confirm accountability, and step in fast when needed.
PQManager
Built to stabilize a failing NCR virtual terminal print queue that would constantly fill and crash printing. After support could not resolve the issue, I created a utility that automatically clears the queue and keeps operations running. It has been running in production for years with zero repeat issues or support calls.
Gping
Originally built as a fast way to verify whether something is actually reachable when systems start acting up. It is currently being reworked into a service-based tool as the project evolves.
GCleaner
Built to recover hard drive space on grocery store registers before they hit the point of failure. Support techs were losing time logging in, hunting through systems, and sometimes missing the cleanup locations that mattered most. I built a discovery-first tool that shows what can be cleaned before the real cleanup script runs, making the process faster, safer, and more consistent. The remaining problem is Windows installer bloat, which needs a separate tool and is already on my list.
CodexSystem / AuditME
An agent operations system I am actively building to keep AI work aligned, auditable, and moving faster with less friction. It tracks tasks, branch state, project truth, agent context, and future work across the repo, while CLI tools and validation scripts help agents stay synchronized and avoid drift. AuditME is the hardening layer inside that system, built to catch annoyances, enforce consistency, and keep multi-agent work usable in the real world.
Sales Order OCR Extraction
A Google Sheets and Apps Script workflow I built to extract sales orders from two completely different source formats and turn them into a clean, structured output. From a custom menu in Sheets, I can choose the correct extraction path for each order source, then export the results into a template-driven sheet formatted exactly how our install process needs it. It is still being stabilized, but the larger goal is to evolve it into a more automatic workflow that also feeds asset tracking, inventory visibility, and CRM build-out.
Principles
How I build
Real use beats theory
If it does not solve a real problem in a real environment, it is not finished.
AI is leverage, not autopilot
I use AI to move faster, but pressure-testing and real-world fit still matter.
Systems should reduce friction
Good tools remove repeat pain, improve clarity, and hold up under pressure.
Contact
Want to see more?
Public repos are linked above. For collaboration or portfolio questions, GitHub is the best first stop.