SUKI SONG I build the AI systems your business actually runs on. Claude Code, MCP servers, integration.
Most AI projects stall because nobody owns the delivery. I build the integration and run the rollout: scope, cadence, gates, handoff. You get a working system, not a pilot.
Your rollout is run by a certified program manager.
Anyone can wire up an API. The reason AI projects die is delivery: no scope, no cadence, no owner, no definition of done. I ran engineering and analytics delivery at Meta as the program manager and Scrum Master for 34 contributors across four teams, and before that at Porsche. Your rollout gets that same cadence.
- PMP Project Management Institute, 2023
- PSM II Scrum.org, 2024 — top 5% globally
- PSPO II Scrum.org, 2025
-
Claude Code for your team
Set up properly, with the guardrails: what it can touch, what it cannot, and how the work gets reviewed before it lands.
-
An MCP server for your business data
A safe, tested door between your systems and an AI assistant. Read-only where it should be, audited where it matters, with the tests to prove it.
-
AI wired into the tools you already run
Integration into your actual stack, not a parallel one nobody opens. If the answer is that you do not need this, I will say so.
-
The plumbing underneath
Automation, dashboards, and the data pipelines that make the output trustworthy. Most of the value in an AI project is the boring layer beneath it.
-
Delivery, run properly
Scope, cadence, gates, and a handoff you can maintain without me. The system is yours, documented, with its tests.
-
Call
Fifteen minutes on what you actually do all day, and where the time goes. Free. If AI is not the answer, this is where I say so.
-
Scope
One workflow, written down: what it touches, what done looks like, and what it is worth. You get the number before any work begins.
-
Build
I build it against that scope, on a cadence you can see, with gates that have to pass before anything lands.
-
Handoff
The system is yours: documented, tested, and maintainable without me. Staying on afterward is an option, never the plan.
I hired Suki to help install AI workflows into my day to day operations and provide ongoing technical support, and the experience was fantastic from start to finish. She’s knowledgeable, friendly, and clearly loves what she does. She took the time to understand exactly what I needed and designed a workflow that actually fit how I work, not just a generic solution. If you want something done right the first time, Suki is who you call. I’d hire her again without a second thought — highly recommend.
Run it yourself.
A live MCP server I built and operate, public and free to hit. Five tools, fifty unit tests, twenty-four live protocol checks against the running server. No signup, no sales call, no form.
Open the demo server{
"tool": "check_availability",
"input": { "date": "2026-08-13" },
"output": { "slots": 3, "next": "09:00" }
} - SOURCE
- Live server, built and operated by me
- METHOD
- MCP over Streamable HTTP, Cloudflare Workers
- STATUS
- 50/50 unit tests, 24/24 live protocol checks
The site you are reading is gated by twenty-one automated checks, and every one of them was proven to fail before it was allowed to pass. That is the standard your system gets.
What you get, and what I will tell you straight.
- A working system, not a demo that only runs on my machine.
- Tests you can run yourself, and documentation you own.
- A straight answer about whether AI is the right tool for the problem.
- If AI is not the answer, I will say so on the call. It will not fix a process that is broken without it, and I would rather tell you that than sell you a system.
What is an MCP server, in plain English?
A safe, documented door between your business data and an AI assistant. Instead of pasting things into a chat window, the assistant can look up your real information through a door you control: you decide what it can see, what it can change, and what gets logged.
Do I actually need one?
Often not. If your team asks the same handful of questions of the same system, an MCP server pays for itself. If you are experimenting, a lighter integration is usually the honest answer, and I will say so rather than sell you a server.
What does it cost?
It is scoped on the call, and you get the number before any work begins. I do not quote a fixed price before I understand the system: the work varies too much to list a number honestly, and I would rather give you something real than a range you cannot plan against.
Who does the work?
I do. When a build needs deeper engineering than one operator should take on, I bring a vetted bench I have worked with, and I stay accountable for delivery.