Azza Arriza — UI/UX & Product Design

MySpark mentor dashboard with a commitment streak and a role skill map

MySpark mentee dashboard with commitment completion and mentoring sessions

MySpark live mentoring session recording with voice activity and mentee commitments

MySpark AI-generated mentoring report with an overall summary and key insights

MySpark

Paragon's internal mentoring app. It holds the whole relationship, the sessions, the commitments people make in them, and the progress that shows they're actually moving, so mentoring stays something you feel, not something you file.

Context

MySpark is Paragon’s internal mentoring app, built at Rakamin and used across the company for people development. Mentoring here isn’t a casual thing on the side. It runs as a real program across roughly 5,000 people, with three kinds of participant: mentors, mentees, and people developers, the specialists, often external, whose whole job is to mentor. And the roles aren’t fixed. The same person can be a mentor to someone junior and a mentee to someone more senior, at once.

Traditionally that mentoring happened the simplest way: a mentor and a mentee sitting down together, offline, notes taken by hand. I took the project over from the previous designer in January 2026 and was the sole product designer on it through June.

The problem

The app exists because that offline conversation came with a hidden tax. Taking notes by hand meant the mentor was doing two jobs at once: listening to the person in front of them, and writing down what was said so there’d be a record afterwards. Attention split between the two, and neither got the mentor’s full focus.

And once the session ended, nothing held it. Mentors had no clean view of their schedule or their past sessions. Mentees had no reliable way to keep the things they’d agreed to work on. The commitments made in one session were the whole point of the next one, and they were living in someone’s handwritten notes.

What I did

Made the session report draft itself from the transcript, so the record writes itself while the mentor stays present, and built the review step as a Word-like editor so correcting it is familiar, not a new chore. The transcript stays the source: fix a mishearing there and the report regenerates, edit the report and the transcript stays put.
Added a profile swap so the same person can be a mentor in one pairing and a mentee in another without juggling two accounts, since that’s how mentoring here actually works.
Reordered each role’s dashboard to lead with what that person actually comes to check, and added status and type badges to the session and commitment lists, turning a wall of near-identical rows into something you scan by shape and color.
Designed the session to run online as well as offline in one flow, so a pair split across offices isn’t left out and neither mode is a half-supported afterthought.
Built skill validation into the session itself, so a mentor assesses a mentee against defined skill levels while the context is fresh, not on a form filled in afterward.
Made the tutorials guided walk-throughs, one per role, taking a newcomer through running a real session rather than reading about it, with onboarding up front to frame what MySpark is before the dashboard.
Built the new screens from the existing component library rather than one-offs, so what I added stays consistent with what was there.

What moved

~5,000

Internal users

A reach number, not a results number. It counts the people who showed up in the backend running real mentoring through MySpark, which proves the app shipped and got used. Whether it made anyone a better mentor is a longer study than I had.

3

Roles designed for

Three audiences with genuinely different jobs, served by one product instead of three bolted together. Each of mentor, mentee, and people developer gets a real fit, without the app going generic to cover all three at once.

Surfaced

Growth on the homepage

An internal mentoring tool's failure mode is becoming admin, a box you tick and resent. I put progress on the dashboard itself, so the first thing you open to is what you've committed to and how far you've moved. Using it reads as improving, not reporting upward.

Open to full-time or freelance
UTC+7 · async-friendly

What are you working on?

Whatever it is, a role, a project, or something you're still shaping, I'm happy to talk it through. Best case, it turns into work I'm glad I took on. Worst case, we find out I'm not the right fit and you're out twenty minutes.

Book a call

azzauiux@gmail.com LinkedIn

Epi-Go daily puzzle asking the reader to transliterate an aksara inscription

Epi-Go home with level progress, lesson tracks, and the daily challenge

Epi-Go transliteration quiz reading a single aksara character

Epi-Go picker of ancient inscription and manuscript scripts to learn

Epi-Go

A learning app for Indonesian epigraphy. Fewer than 200 people can read these stone inscriptions, so I built a way in that doesn't need one of them in the room.

Context

Almost half of Indonesia’s ancient history is written in Prasasti, stone inscriptions carved in scripts most Indonesians can’t read. Fewer than 200 people in the country can actually read them. So the story of where you’re from sits in plain sight, in museums and at temple sites, locked behind a script that almost nobody alive still knows.

Epi-Go is my attempt at a way in. I found this gap while I was at university, and it stuck with me. The history is right there, and the only thing standing between a curious person and reading it themselves is that the knowledge lives with a couple hundred experts. So I built a learning app that doesn’t need one of them in the room.

Background and Problem

Learning a script almost nobody reads is hard in a way that learning French isn’t. The inscriptions themselves aren’t the problem. You can stand in front of a Prasasti at a museum, or pull a photo of one up online. The wall is that you can’t read what it says, and almost nobody can teach you. There’s no evening class for it, no app already built for it, no friend who happens to know it. The knowledge exists and the artifact is right there. The way in is the missing piece.

So before designing anything, I went to find out how people actually learn a non-Latin script from a cold start. I ran a questionnaire on what people knew about epigraphy and whether they’d ever learned a non-Latin script, then interviewed 30 of them about how they got through it. Three findings came back clear enough to build on. Start from the simplest piece and climb. Gamify the parts that traditional study makes a slog. And let people learn alongside each other instead of alone.

What I did

Built the learning path to climb from the simplest piece up, so someone who’s never seen the script starts by recognizing one character and ends up reading a line, instead of drowning on day one. I added a helper table they can pull up mid-lesson, but capped how many times, so it’s a hint and not a crutch.
Gamified the slog. The parts of learning a dead script that are pure repetition got turned into something with progress and small wins, so the boring-but-necessary drilling stops feeling like punishment.
Built a place for learners to gather, share where they’re stuck, and talk through the script together, because the research was clear that nobody learns this alone for long.
Added a leaderboard and a streak counter, kept healthy on purpose, so the competition motivates a daily habit instead of scaring off the people who are behind. The leaderboard sits on the homepage for a quick glance, and the streak rewards showing up every day.

What moved

3.8%

Accuracy gain

This isn't a jump from bad to good. People were already reading a Prasasti at 81.4% accuracy, and gamified learning moved them to 85.2%. Late gains on an already-high number are the hard ones to win, so those few points mean more than the size makes it sound.

60%

Preferred gamified learning

Most people who tried both said they'd rather learn this way than the conventional drill. That's a preference, not a test score, but it's the thing that decides whether someone opens the app a second time. For a subject people usually quit, wanting to come back is most of the battle.

89.25

SUS score

System Usability Scale, graded A, in the range they call excellent. This is the number I trust most in the project. It says the app itself got out of the way, so the difficulty was the script, which is the hard part, and not the interface, which shouldn't be.

Open to full-time or freelance
UTC+7 · async-friendly

What are you working on?

Whatever it is, a role, a project, or something you're still shaping, I'm happy to talk it through. Best case, it turns into work I'm glad I took on. Worst case, we find out I'm not the right fit and you're out twenty minutes.

Book a call

azzauiux@gmail.com LinkedIn

Redesigned Coffee Works Express homepage with the featured machine hero

Coffee Works Express product detail for the WEGA Nova espresso machine

Coffee Works Express

A coffee-equipment site for Sydney and Melbourne. One catalog serving a buyer researching a machine and an owner who needs one serviced, routed so neither has to hunt.

Context

Coffee Works Express has imported and serviced commercial coffee machines since 1998, selling to cafés, roasters, offices, and home baristas out of Sydney and Melbourne. It’s a real specialist business. Josh, who runs it, knows these machines down to the group heads and the grinder types.

Two kinds of people land on the site for opposite reasons. One is a new customer trying to work out which espresso machine to buy. The other already owns one and needs it serviced. The whole design problem was serving both without the site pulling itself in half, and doing it in a domain where the person browsing usually knows far less than the person who built the catalog.

What they hired me for

Nothing was broken. The old site just hadn’t kept up. It worked, but it looked dated next to where coffee-equipment sites had gone, and Josh wanted it modernised. So this was a fresh redesign, not a rescue. He came in with a concept and a clear sense of his own product, set a loose direction, and left room to argue with it. My job was to develop that into a working site.

The real constraint wasn’t anything on the page. It was the domain. Josh knows coffee machines down to the group heads and the grinder types. Someone landing on the site to buy their first machine knows none of that. The job was to build an interface that carried his knowledge across to a person who doesn’t have it, without making either of them feel talked down to.

What I did

Designed the homepage around a featured-product showcase, because a first-time buyer who can’t tell one espresso machine from another needs the client’s own recommendation to anchor to before anything else. It’s the one screen doing the most work for the person who knows the least, so it leads.
Threaded a service path through a buying-focused site, so a customer who already owns a machine and needs it repaired reaches a real booking flow instead of a dead contact form.
Reordered the support flow so clicking a category, parts, sales, or service, sets the request context, instead of making the customer describe their problem from a blank start.
Built a data-driven template system, 6 templates rendering 40+ machines across 9 brands, so the catalog scales by adding data, not by a designer redrawing a page for every new product.
Reviewed the developer’s live build against the designs and marked up where the shipped site had drifted, so what launched matched what was designed.

What moved

40+

Machines, six templates

6 page templates carry the whole catalog, 40+ machines across 9 brands. The system scales by adding data, not by redrawing a page each time a new machine lands. A count from the live site, not a target.

2

Opposite intents, one structure

A first-time buyer researching a machine and an existing owner needing one serviced want opposite things from the same site. One structure routes both without either having to hunt, and without the page splitting in two to do it.

Owned

From client brief to production

Josh came in with a concept. I took it from that brief through to the shipped site, developing the design, then reviewing the developer's live build and marking up where it had drifted. Not one slice of it, the whole thing.

Open to full-time or freelance
UTC+7 · async-friendly

What are you working on?

Whatever it is, a role, a project, or something you're still shaping, I'm happy to talk it through. Best case, it turns into work I'm glad I took on. Worst case, we find out I'm not the right fit and you're out twenty minutes.

Book a call

azzauiux@gmail.com LinkedIn

RODE vertical SOP video feed with likes, comments, and an attached PDF

RODE new-post composer with a video cover, topic, and PDF attachment

RODE audience picker targeting posts by division and department

RODE

Paragon's SOP feed, built like TikTok. Nobody read the change-management documents, so I turned them into a short-video feed people actually scroll.

Context

RODE is Paragon’s internal SOP feed, built at Rakamin. When a big company changes how something is done, a new safety procedure, a new process, a policy update, that change has to reach the people who actually carry it out. Paragon’s way of doing that was the same way most companies do it: long documents, sent around, meant to be read.

The catch is who’s on the other end. The people who need an SOP are spread across the whole org, from directors down to the specific function on the floor, and a change rarely applies to all of them at once. A procedure meant for one department shouldn’t land as noise for everyone else. So the real shape of the problem wasn’t just “make the documents nicer to read,” it was getting the right change in front of the right slice of a large, layered organization, in a form they’d actually take in.

Problem

They read the documents. That’s the part everyone gets wrong. The problem was never neglect, it was shallow reading. An SOP is a wall of text describing something that changed, and the people it’s for are busy doing the job the SOP is about, so they skim. They catch the headline and miss the half that matters. A lot of them told me they’d just paste the whole thing into an AI, read the summary, and move on.

The trouble is that a summary of a procedure isn’t the procedure. Someone can come away sure they’ve absorbed a change when they’ve only skimmed the shape of it, and the old format had no way to tell the two apart. Read and understood looked identical from the outside. That’s the gap RODE had to close.

What I did

Turned the SOPs into a vertical video feed people scroll like TikTok, so a change someone used to skim as text is now something they watch through to the end. The PDF stays attached and downloadable, so the feed is where you learn it and the full document is there when you need the exact wording.
Built content targeting down the org hierarchy, four levels from directorate to function, so a procedure meant for one team reaches that team and doesn’t land as noise for everyone else.
Designed the like, view, and comment counts to sit separately after working through it with the PM, because a merged count read as if tapping it did something it didn’t, and separating them made each number honest about what it measured.
Made the caption expandable and the PDF download bar always visible, so someone can skim the gist in the feed or pull the source document without leaving the video, two ways into the same SOP.
In Phase 1.2, added edit and delete with deliberate friction, soft-delete and a lock on the uploaded video, so a creator can fix a mistake without silently rewriting an SOP people may have already acted on.

What moved

4 levels

Targeted by org unit

Content targets down the org hierarchy, directorate to division to department to function. A real count from the PRD, so a procedure reaches the team it is for instead of everyone.

5 epics

Owned end-to-end

Sole designer across all five Phase 1 epics, from login through the feed, social interactions, and profile. Owned the whole MVP surface, not one slice of it.

2×

Time spent learning

The short-video feed kept people with the material instead of bouncing off it. Early cohorts spent roughly double the time learning an SOP. A modelled read, a strong signal, not a measured result.

Open to full-time or freelance
UTC+7 · async-friendly

What are you working on?

Whatever it is, a role, a project, or something you're still shaping, I'm happy to talk it through. Best case, it turns into work I'm glad I took on. Worst case, we find out I'm not the right fit and you're out twenty minutes.

Book a call

azzauiux@gmail.com LinkedIn

Redesigned Stryve Finance homepage with the lending-advice hero

Stryve equity-release landing page for accessing home value

Stryve insights blog with a featured loan-types playbook

Stryve blog category browser across loan topics

Stryve client testimonials wall with Google review cards

Stryve Finance

A Sydney mortgage brokerage site I still run. One audit turned into twenty-plus pages and the component library that keeps them all speaking the same language, nine months and counting.

Context

Stryve Finance is a Sydney mortgage broker working across fifty-plus lenders. For most of the people who land on the site, a mortgage is the largest financial decision they’ll ever make, so the whole job of the page is to earn enough trust that a stranger picks up the phone.

I took the site over from a previous designer in November 2025. The homepage was already live. Almost everything else, the loan product pages, the suburb template, the business loan pillar, I designed from scratch as the site grew.

Problem

The site wasn’t earning that trust. Across 9,553 sessions, people scrolled 37% of the way down the homepage and left, 1.43 pages a visit, gone before they reached the phone number. The page had plenty on it. It just wasn’t arranged to make anyone call.

The audit was what Stryve hired me for, and it surfaced three things worth fixing. Components had drifted out of sync as the site grew, so the same button or card rendered differently from one page to the next. The hero wasn’t selling. And the FAQ led with the questions that were easy to answer instead of the ones people were actually anxious about.

What I did

Reordered the homepage around an argument it didn’t have. It opened by listing loan options before it had earned any reason to trust Stryve, so I resequenced it to build trust first, then show what separates Stryve from other brokers, then proof of work, then testimonials. I argued for the order in the file at the time, it’s the difference between a page that lists things and a page that walks someone toward calling.
Reordered the FAQ to lead with the questions people are actually anxious about, instead of the easy ones, so the section that should close the deal stops warming up on the wrong questions.
Built the component library as the site grew, so the drift stopped. One button set now runs 176 instances across 13 pages, which is the real test, the library gets used instead of worked around.
Designed one dynamic template for the suburb pages, filling from data like median prices and clearance rates, so Stryve could rank for dozens of locations without hand-building a page each time.
Designed the newer pages from scratch as the site grew, the loan products, the profession-specific pages, the business loan pillar, so the twenty-plus-page site holds together instead of reading like parts bolted on.

What moved

+23%

Conversion

The old page buried the phone number under everything a visitor didn't ask for. I cut it to the parts that make someone call. Modelled from the client's closing range after launch, so I'd call it an estimate I stand behind, not a tracked number.

20+

Pages designed

Homepage, loan products, suburb pages, the business loan pillar, calculators. Most of them designed from scratch as the site grew past the original audit.

Trust

Page resequenced

The page had no argument, people scrolled 37% and left. I reordered it around a sequence I could defend, build trust, then show what sets Stryve apart, then proof, then testimonials. A decision visible in the file, not a measured funnel change.

Open to full-time or freelance
UTC+7 · async-friendly

What are you working on?

Whatever it is, a role, a project, or something you're still shaping, I'm happy to talk it through. Best case, it turns into work I'm glad I took on. Worst case, we find out I'm not the right fit and you're out twenty minutes.

Book a call

azzauiux@gmail.com LinkedIn

I Make OVERWHELMINGCONFUSINGOVERWHELMING stuff feel simpleAnd obvious

Selected work

MySpark

Epi-Go

Coffee Works Express

RODE

Stryve Finance

What the work moved

What I do

UX & PRODUCT DESIGN

UX AUDITS & CRO

BRAND & WEB DESIGN

Where I've worked

UI/UX Designer at WebRocket

Freelance UI/UX Designer

Product Designer Intern at Rakamin

What are you working on?

I Make CONFUSING
stuff feel simple
And obvious