PORTFOLIO — 2026
WORK / 03
PRODUCT · IN DEVELOPMENT
A short, structured journal for boxing classes and solo practice. Each session is recorded in the same frame, so lessons can be recalled, linked and built on.
INTERFACE DESIGN
01 · OVERVIEW
Boxing classes move fast. A coach’s correction is clearest right after class and fuzzy by the next one. BTJ is an iPhone journal that keeps the coach’s words, turns them into one thing to practice alone, and records what changed. I’m also its first user: every record so far comes from my own classes. I own the product — requirements, decisions and hands-on acceptance — while AI does most of the implementation inside a workflow I set up.
02 · PROBLEM
01
A correction is precise in the moment and vague a week later. Paraphrasing it changes what the coach meant.
02
Alone, it’s easy to practice something next to what the coach actually said.
03
Without the same frame every time, one session can’t be compared with the next.
THE DESIGN QUESTION
How do you keep a coach’s words intact, turn them into one focus, and still let the record be corrected — without rewriting what happened?
03 · THE SESSION
Every session follows the same path, and the app only saves at points that mean something.
STEP 1 · COACHING
Right after class: what you worked on, the coach’s advice in their own words, and what improved or felt difficult. All three are required.
NOTHING SAVED YET
STEP 2 · PRACTICE
Edit a draft built from your notes while the coach’s original words stay visible. Pick one practice focus.
FOCUS SAVED · SESSION CREATED
Drills, context, a 1–10 consistency check, and when it worked or didn’t.
HELD UNTIL REFLECTION
STEP 3 · REFLECTION
What changed, what to check with your coach, and the next focus. Saving completes the session.
SAVED TOGETHER · SESSION COMPLETE
THE RECORD
Technique progress lists every session oldest-first, so changes show up over time — and it is where a backup export starts.
04 · LIVING WITH IT
A journal kept for months needs ways to fix, remove, read and take out what’s in it — without surprises.
Edit one section at a time
Original notes, coaching summary, practice and reflection are edited separately. Changing one won’t rewrite the others; dates, status, conditions and scores stay fixed.
Delete on purpose
Permanent deletion asks you to type the session number and says exactly what goes — and that exported files and backups don’t.
Switch language, keep the words
The interface switches between English and 繁體中文. Notes stay as written; nothing is machine-translated.
Take a copy out
Export creates a backup file from Technique progress. If the share sheet closes, the app says it can’t confirm a copy was saved.
WHEN THE APP CAN’T BE SURE
If a save can’t be confirmed, the app shows “Save not confirmed,” asks you to retry the same change, and pauses other actions until it’s resolved. The same rule covers edits and deletion.
05 · DECISIONS
DECISION 01
NOT CHOSEN
Option A · Lock it
+ The record can’t drift after the fact.
− Typos and wrong words become permanent.
CHOSEN
Option B · Allow edits, one section at a time
+ Notes can be corrected without erasing the session.
− Needs rules for what can’t change: dates, status, scores.
Why: Correcting a note isn’t the same as canceling a session, so editing and deletion became separate actions.
DECISION 02
NOT CHOSEN
Option A · Translate them
+ One language on screen.
− Rewrites the coach’s words and can add errors.
CHOSEN
Option B · Keep notes as written
+ The coach’s words stay exactly as recorded.
− Older notes may not match the interface language.
Why: The point of the journal is the coach’s exact words.
DECISION 03
NOT CHOSEN
Option A · Assume it worked
+ Fewer interruptions.
− Silent data loss looks like success.
CHOSEN
Option B · Say so and pause
+ You know to retry before moving on.
− An extra screen in rare cases.
Why: A journal is only useful if you can trust what’s in it.
All three were my calls as product owner; AI proposed and implemented the options.
06 · BUILT WITH AI
The practical problem: AI can move from a suggestion to a “completed” report faster than I can check every technical claim. So the work runs through three gates.
01
Define permitted changes, actions, stop conditions and deliverables. Finishing one task doesn’t authorize the next.
02
Separate local measurements, checked attachments, cited reports and unknowns. Repeating a report isn’t new evidence.
03
I check what the product actually does. Designated reviewers cover technical and security questions within their scope. Missing evidence stays unverified.
An AI response read a request to draft a session-maintenance plan as permission to start building. I restated the drafting scope and settled the open product decision first.
I rejected the first English copy and asked for revisions, then ran seven manual checks myself: bilingual use, input preservation, saved preferences, reopening, original-language notes, and narrow and enlarged layouts.
The roles and gates are documented workflow controls. They are not technical permission isolation, autonomous multi-agent operation or independent security assurance.
07 · ROLE & LIMITS