AI systems audit
I map where AI actually fits your workflow, what is safe to automate, and what will break. You leave with a written plan and a priority order, not a demo.
Accepting clients
AI Systems Consulting
Stop building AI dashboards.
Start engineering business value.
I don't build flashy dashboards that look impressive and change nothing. I bring nearly two decades of engineering discipline to AI: every idea filtered by what is legally compliant, technically achievable, and genuinely worth it to your bottom line. What survives ships specified, tested, and built into your daily workflow to cut hours and costs.
▸ Enter Engine Room
the dial · click to switch rooms
- 18+ yrs
- Engineering
- 35
- Lab-log entries
- spec-first
- Always
- 2026→
- In public
// Services
How I put AI to work
Every engagement runs the way I ran hardware programs: understand the operation, find the pain, propose the fix, then build and implement it. Write the spec, test against it, ship what survives.
UNDERSTAND → DIAGNOSE → PROPOSE → IMPLEMENT
Custom assistants + agents
Assistants and scoped agents wired to your tools and data, with human-in-the-loop guardrails. Built to a written spec, with inputs, outputs, and acceptance checks agreed up front, and run against those checks on your real documents before they touch live data.
Accepting clients
Practice areas · where I'd put it to work first
AI for commercial real estate
Listing collateral, lease abstraction, landlord reporting: the paperwork layer of a deal practice. The relationships stay yours; the production line is where AI goes to work.
Accepting clients
▸ Opportunity menu
AI for dental practices
Front-office and marketing workflows first: communications, content, imaging prep. Anything touching patient records waits for the right agreements and stays on machines inside your practice.
Accepting clients
▸ Opportunity menu
AI for rental management
Applicant paperwork, compliance notices, lease templates, rent tracking. Structured support for anyone running rentals; the tenancy decision stays yours.
Accepting clients
▸ Opportunity menu
Different industry? The method is the same; these three are just where I'd start.
▸ Book a working sessionEquipped with · running the line
Claude
Midjourney
ElevenLabs
Gemini
Accepting clients. Nothing is for sale yet. I do not sell a service until it has shipped for real users, an engineering rule that applies to the business too. That makes early engagements the best terms this practice will ever offer: my full attention, a scope we set together, and a result I need to prove even more than you do.
Book a working session// Work
The proof is the record, not the pitch
No client case studies yet, and I would rather show real ones than borrow someone else's words. What I can show is the work itself: a dated log of what I have built and what broke, kept in public in the Engine Room.
The public record
35 dated lab-log entries, and counting.
A running engineering log of building a personal AI operating system from scratch: the wins, the failures, and what each one taught. It lives in the Engine Room, the technical view for builders and peers.
Client results · reserved
Client results and testimonials go here once engagements ship and the outcomes are real. These are placeholders, on purpose. Better empty than filled with words no one said.
Reserved for a real client quote
Reserved for a real client quote
// About
The engineer behind the AI
- ROLE:
- Staff Engineer Electromechanical
- SECTOR:
- Defense aerospace, avionics packaging
- EDU:
- M.S. & B.S. Mechanical & Aerospace Engineering, UC San Diego
- CERT:
- Systems Engineering, Caltech CTME
- LANG:
- English · Mandarin
- BASED:
- San Diego, CA
By day, I am a W-2 engineer in defense aerospace, where national security rides on engineering done right. Nights and weekends, I am an AI Generalist: the same discipline, pointed at AI for small business. AI is moving fast enough to make good businesses obsolete; the owners with the vision to move first are the ones who stay ahead. That is what this practice is for: custom AI built into the daily SOP they already run.
The engineering half is real. For nearly two decades I have worked in mechanical design, failure investigation, and supplier engineering on hardware where a leak the width of a bacterium is a program-level event. FEA, GD&T, root-cause analysis, and shipping against specifications that do not negotiate.
AI work today looks the way early CAD did: powerful, unstandardized, full of people improvising. Engineering discipline transfers directly: written specs, controlled tests, honest failure logs. That is what I bring.
Selected record
[OK] mechanical_lead · production validation, RF microwave hermetic device
[OK] failure_authority · RCCA / MRB tiger teams · hermetic, weld & solder non-conformances
[OK] trainer · GD&T and FEA courses delivered to a working engineering org
[OK] builder · ERP-integrated scheduling software, 2009: the crossover started early
// Writing
Notes on building AI like an engineer
Plain-language essays on treating AI as a system: what transfers from hardware engineering, where the tools fall short, and how to tell a real capability from a demo.
The first pieces are in draft. They go live here as they ship.
Draft
"Spec-first prompting: write the requirement before the prompt"
Treat a prompt like a drawing you have to sign off on. Write the acceptance criteria first, then work backward to the words.
Draft
Agents need guardrails, not autonomy for its own sake
The interesting question is not how much an agent can do on its own. It is where you put the human, and what the agent is not allowed to touch.
Draft
Running an AI video pipeline like a production line
Script to voice to visuals to a finished cut, treated as stations on a line with quality gates between them, not one heroic prompt.
Want the first posts when they land? Say the word and I will send them.
// Contact
Let's build something dependable
A working session is the front door: bring the workflow that eats your week, and we find out what AI can and cannot do about it. If there is a fit, the next step arrives in writing. Questions and curiosity welcome too; I read everything.
Direct · active
contact@xcraftai.com. Tell me what you are trying to build and what is in the way. If it is a fit, we will scope a working session.
Professional · active
LinkedIn carries the full engineering record, the career the AI work stands on.
Accepting clients. Nothing is for sale until it has shipped for real users; early engagements are where we set the terms together.