The clean version of the story sounds fake.
“I moved from immunology operations into digital asset security.”
That sentence is true enough and still hides the useful part.
The useful part is not that I made some dramatic leap from science to crypto. It is that the habits transferred more cleanly than I expected.
In immunology and cancer research operations, sloppy process has consequences. Samples need identity. Protocols need version control. Deviations need documentation. Handoffs need clarity. People need to know what changed, who touched what, and what evidence supports the conclusion.
Digital asset security is a different world, but the posture rhymes.
Who has access?
What is the source of truth?
What happens if the normal person is unavailable?
Where is the recovery path?
What did we verify, and what are we just assuming?
That is the bridge.
I learned to respect boring procedure
In research operations, boring procedure is not decorative.
Label the thing.
Record the thing.
Follow the protocol.
Write down the exception.
Do not silently improvise when the handoff depends on someone else understanding what happened.
That mindset is easy to underestimate until you work around systems where a small mismatch can waste weeks or ruin trust in the output.
Digital asset custody has the same problem in a different costume.
A wallet setup that only one person understands is fragile.
A seed phrase without a recovery procedure is fragile.
A passphrase remembered but not recoverable is fragile.
A beneficiary plan nobody updated is fragile.
A transaction review process that changes depending on mood is fragile.
The tools are different. The failure shape is familiar.
Verification became the through-line
The phrase I keep coming back to is: in bitcoin, you do not trust, you verify.
I do not use that as a bumper sticker. I use it as an operating habit.
In research, you do not trust a result because it sounds plausible. You check the source material, the method, the sample, the run, the notes.
In AI-assisted building, I do not trust the model because it sounds confident. I check the diff, run the build, open the app, inspect the mobile layout, and read the copy out loud.
In retirement modeling, I do not trust the chart because it looks clean. I want the formula trace.
In custody workflows, I do not trust a setup because the hardware wallet screen looked official. I want the recovery path, the dry run, the written procedure, and the transaction review habit.
That sounds repetitive because it is.
Verification is the job.
The AI part did not erase the operator part
AI helped me build things I would not have built by hand.
Mandarin Deck started because I moved to Hong Kong and wanted an HSK 3.0 tool I would actually use. The first version was a single HTML file. Then it grew into a PWA with Supabase sync, Google OAuth, native-speaker audio, reading mode, and a service-worker escape hatch after I learned how badly phone updates can trap themselves.
RetireCalc started because someone in my family asked which retirement account to draw from first. That became a browser-only planning tool with ACA cliffs, IRMAA lookbacks, Social Security taxation, Roth conversion windows, synthetic profiles, and a formula trace.
This site started as a portfolio. It is becoming more of a proof trail: what I built, what broke, what I tested, what I still do not claim.
AI made the implementation possible.
Operations made the work survivable.
The model can write code. It cannot decide what evidence should count. It cannot know when the wording overclaims. It cannot feel the difference between a chart that looks finished and a model that someone can inspect.
That is still my job.
Digital asset security is mostly human systems
People like to make custody sound like a hardware decision.
Which wallet?
Which metal backup?
Which multisig coordinator?
Which exchange?
Those matter. They are not the whole system.
The real questions are messier:
Who should be able to move assets?
Who should not?
What happens if a device is lost?
What happens if someone dies?
What happens if someone is pressured?
What happens if the person who set everything up cannot explain it?
What happens if the family finds the seed phrase but not the passphrase?
What happens if the instructions are technically correct but emotionally impossible to follow during grief?
That is why I describe my lane as custody architecture, security workflows, transaction/custody operations, inheritance protocols, controls, and proof.
Not trading. Not portfolio management. Not taking custody.
Architecture and procedure.
The non-linear path helps
For a while, I thought not having a traditional CS background was a weakness.
Sometimes it is. I still hit syntax problems that a trained developer would not hit. I still need tools, tests, and review loops around code I did not grow up writing.
But the non-linear path also helps.
I am comfortable being the person who asks, “What is the actual procedure?”
I am comfortable making a checklist.
I am comfortable saying, “This is not proven yet.”
I am comfortable writing down the boring thing everyone assumes they will remember.
Those habits are useful in digital asset security because the failures are often boring until they are catastrophic.
Lost phone.
Wrong recovery email.
Dead hardware key.
Unclear passphrase.
Outdated beneficiary.
No restore test.
One person knows the whole setup.
None of these require a genius attacker. They require neglect.
What I am trying to build now
The through-line is clearer to me now than it was while I was living it.
Mandarin Deck is about turning a learning problem into a system I can actually use every day.
RetireCalc is about turning a family planning question into a model with visible assumptions.
Fortress21 is about turning digital asset holdings into custody workflows that can survive mistakes and succession.
The Lunatechs events are about helping people open the tool, hit the first error, and keep going.
The Bullish work sits in the same broad lane: digital asset security, operational discipline, and systems where proof matters.
I am not trying to pretend all of this was obvious from the beginning.
It was not.
The path only makes sense when I look backward: immunology operations taught me to respect evidence, handoffs, and procedure. AI gave me a way to build tools around those habits. Digital asset security gave those habits a sharper place to land.
That is the article I would have wanted to read earlier.
Not “how I reinvented myself.”
More like: the skills you think are adjacent may already be the core skill, once the stakes change.