Current-state audit
Map current tools, repeated decisions, manual handoffs, data sources, and the points where work slows down or becomes inconsistent.
AI consulting
I design practical AI systems that remove repeated work, not add another layer of complexity.
Most AI work fails because people start with the model instead of the workflow. I help map where AI can actually carry useful work, where it should stay out of the way, and how to connect it to the tools a team already uses.
AI consulting
What this can cover
Agentic workflow design
Design agent-like flows with clear inputs, boundaries, checks, fallback paths, and human approval where the system should not act alone.
Measured rollout
Define acceptance criteria, logging, error handling, ownership, and test cases so the first version can be trusted before it is expanded.
Toolbox
The AI tools I actually work in
Not a buzzword list. These are the models and platforms I set up, compare, and use for real work every day.
Claude
Certified through Anthropic Academy across a broad part of the ecosystem. My primary tool for coding, agentic work, and written content.
Codex
Have used Codex since it first shipped, for fast code generation and scoped automation tasks.
GLM, DeepSeek, Kimi
Delegate scoped tasks to GLM, DeepSeek, and Kimi when the job fits them better, and cheaper, than the larger models.
Local LLMs
Set up small and larger local models myself and know which model fits which task type when the data needs to stay on the machine. See my stance on local AI.
Practical AI scope
How I would use this
Small teams with messy operations
Best for founders or teams that already have repeated work happening in email, spreadsheets, forms, CRMs, docs, or internal tools.
Typical outcomes
A clear workflow map, a realistic AI architecture, prioritized build steps, and a first version that reduces manual work without hiding risk.
Delivery format
Usually starts as a focused scope session, then becomes either a build plan, a prototype, or a small implementation sprint.