Skip to content
Lahiru Paranavithana

Product designerBased in Australia

Lahiru Paranavithana, product designer

L

Lahiru

At my desk with Revox

Coldplay on, deep in Figma

An illustrated office: Lahiru sits at a desk with the Revox team, wearing headphones and listening to Coldplay, while teammates walk between the desks, meeting room and lounge.

What I do

  • Design systemsMain focus
  • Product design
  • Accessibility

I design experiences as intuitive to use as they are beautiful to see, and the design systems that keep them consistent. Right now I’m with the Revox team; that’s me at my desk, click to say hello.

Kind words from people I’ve worked with

1 / 7

“Lahiru and I worked together for six months in the same team. During this time, I have gotten to know him as a creative and skilled UX designer who was an invaluable asset on all projects our team worked on. From small graphics to large and detailed web pages, Lahiru always delivered great and timely work with lots of attention to detail. I would recommend Lahiru to anyone seeking a creative UX designer to join their team.”
Valeria P.Worked with Lahiru on the same team, March 2023
Read them on LinkedIn

On repeat at my desk: Coldplay

Selected work

Case studies

SaaS, banking and language learning. The ones under NDA need a password; ask me and I’ll share it.

What I focus on

Systems
first.

Good products feel calm because the decisions underneath them are consistent. I spend most of my energy on the system, so every screen built on it starts out accessible and on-brand.

Main focus

Design systems

The craft I most want to master. I build foundations, tokens and components that let teams ship consistent, accessible UI faster, like the Humanised system I built from scratch. Try its primary button.

Token
action.primary.bg
Value
#0052EA

Built in, not bolted on

Accessibility

I check contrast, focus order and screen-reader flow while I design, not after. Currently learning with Intopia. Drag to see where text stops passing.

Your pay run is ready to review.

13 October

  • AA body text (4.5:1)Pass
  • AA large text (3:1)Pass
  • AAA body text (7:1)Fail

End to end

Product design

Discovery, flows, prototypes and testing with real users. Mostly SaaS products, websites and iOS apps across fintech, HR & payroll and edtech.

Domains
Fintech, HR & payroll, edtech
Platforms
Web apps, websites, iOS
Learning
Front-end code, to close the design–dev gap

Reading list

Research I design by

Articles I’ve read and think are worth following if you design for learning.

  • Burr Settles & Brendan Meeder, Duolingo, ACL 2016

    A trainable spaced repetition model for language learning

    How Duolingo predicts when you’re about to forget a word, and what that means for designing practice schedules, progress meters and feedback.

    Read the paper (PDF)
Explore the paper’s ideas interactively
0%50%100%0714 days

Memory fades on a curve

Recall chance halves every half-life: p = 2^(−Δ/h). Duolingo’s skill strength bars are this number. Drag to watch a word go stale.

What I take into designShow progress honestly: strength should decay when practice stops, not stay gold forever.

Practising manteau (coat):

  1. right
  2. wrong
PimsleurFixed schedule, ignores answers
Next review in26 sec
LeitnerDoubles when right, halves when wrong
Next review in1 day

Fixed schedules vs adaptive ones

Pimsleur reviews every word on the same timetable. Leitner adapts, doubling the gap when you’re right and halving it when you’re wrong. HLR learns these weights from data instead.

What I take into designPersonalise practice: an easy cognate and a tricky word shouldn’t share one schedule.

  • visite+0.94
  • suis+0.47
  • trou+0.05
  • dessous-0.06
  • ceci-0.45
  • fallait-0.91

Harder, decays fasterEasier, lasts longer

Some words are simply harder

The model learned a weight for each word. Cognates and short, common words last longer; irregular forms, participles and rare words fade faster.

What I take into designContent design matters: the hardest grammar needs more frequent, better-explained practice.

Prediction error (MAE), lower is better

  • HLR0.128
  • Average guess0.175
  • Logistic regression0.211
  • Leitner0.235
  • Pimsleur0.445

Daily retention after switching to HLR without word tags

Any activity
+12.0%
Lessons
+1.7%
Practice
+9.5%

Better predictions, more learners

HLR cut prediction error well below every baseline. A later version without per-word weights fixed words that decayed unfairly fast, and lifted daily retention.

What I take into designTrust is UX: when students said meters felt wrong, fixing the model brought them back.

Dribbble

More on Dribbble

Smaller explorations, components and UI details. Hover to pause, click to open the shot.

See my Dribbble profile

Contact

Let’s talk.

Need a creative co-pilot, or just want to say hi?

lahirump@yahoo.com

About me

I’m currently working with the Revox team. My experience spans fintech, HR & payroll and education, from Seylan Bank’s mobile app to Humanised’s payroll platform and Accented AI’s personalised language learning.

I specialise in SaaS products, websites and iOS apps, with design systems as the area I most want to master. Alongside work I’m learning accessibility with Intopia, design systems through Dan Mall’s Design System University, and code to bridge design and development. This site is part of that.

I love working with teams to turn creative ideas into intuitive, useful products.

Designed and built by Lahiru, 2026

Back to top