The challenge

The first screen after login did none of the work. Hello Coach is an Australian platform matching people with accredited coaches across career, wellbeing, relationships and personal growth. The production dashboard was a card for your coaching team, a search box and a Life Tracker summary. Being coached happened on other pages.

New members had no first step, and returning members couldn't see their next session. The most common task, checking when your next session is and joining it, meant leaving the page built to be home. The Life Tracker, the platform's signature idea, showed completion percentages with no route from the numbers to an action.

BeforeAfter

My role

Product designer on the dashboard redesign:

• Audited the member journey to decide which tasks belonged on the dashboard
• Designed the onboarding, coach discovery, session and progress modules, with every state from new member to active member to empty
• Built the screens as a documented source of truth for engineering handover

The process

Guidance before features

New members get three steps, in order, on the page where they happen. Pick from your curated coaches, add coaches for other areas of life, book. The guidance is dismissible: it teaches the journey once, then gets out of the way.

Discovery where the intent is

Finding a coach moved from a bare search box onto the dashboard itself. Natural-language search ("what would you like help with?"), deeper filters and tabs for the eight life areas: growth, prosperity, career, wellbeing, relationships, identity, mindset, connections. Coach cards show what members decide on: the face, a line in the coach's own words, experience and location, with booking one click away.

Sessions as the centre of gravity

Each session row leads with the action its status calls for. Starting now says Join, upcoming offers its prep form, finished asks for a rating, cancelled says so plainly. The most common reason to open the app became a zero-navigation task.

Progress you can act on

The Life Tracker™ became a wheel you can act from. Eight life areas with per-area progress, sitting beside goals and your coaches, so noticing a gap and doing something about it happen in the same glance.

The outcome

The dashboard shipped as the daily home of the coaching journey, not a landing page in front of it. New members get a guided path to their first session, returning members join, prep and rate without leaving, and the Life Tracker went from a report to a starting point.

The challenge

AI video was the fastest-moving space in the product. Veo, Seedance, LTX, Kling and more arrived monthly, each with its own durations, dimensions, frame rates, audio support and reference inputs. Leonardo's position depended on shipping them fast: 25+ models across video and image in a single year, in partnership with Google, OpenAI and ByteDance.

Video had started as a modal with one model in it, and every launch strained it further. Each model's controls created new edge cases: unsupported durations, dimensions that conflict with aspect ratios, audio on some models only, costs that change with settings. Handled naively, every model would need its own UI and users would relearn the product with each launch.

BeforeAfter

My role

Lead designer on video generation through this evolution:

• Defined how differences between models are abstracted in the interface, so choosing one never requires reading a spec sheet
• Evolved a single-model modal into a dedicated creation surface without breaking existing users' muscle memory
• Designed how options behave when a model doesn't support them, and aligned design, engineering and model teams as new capabilities shipped

The process

A modal can't hold twenty models

The modal grew into a dedicated surface: pick a model first, then the controls adapt to what it can do. The structure stayed stable (prompt, model, settings, references) so existing users kept their muscle memory while the surface underneath became extensible.

The compatibility problem: references, start frames, end frames

The hardest differences weren't sliders, they were inputs. One model took an image reference, another a video reference, some a start frame, fewer an end frame. Adding a start frame could change which durations were offered. Twenty-five models multiplied by every input type is a compatibility matrix no user should have to think about.

So the interface carries the matrix instead of the user. Every input stays discoverable, each model enables only what it can honour, and anything incompatible shows as disabled with a plain-language reason. Switching models keeps your prompt and references and never throws work away. For new input types like start frames, I designed the onboarding that introduced them.

Rules, not exceptions

Controls follow rules against a model's capabilities, not hand-designed quirks. Unsupported options appear disabled with a tooltip, never hidden, never a dead end. Locked sliders keep their footprint so the layout doesn't jump. Costs update as duration, resolution and audio change. Engineering resolves each new model against the same rule set instead of waiting on bespoke design.

Launch cadence as a design constraint

With launches landing monthly, repeatability was the craft. Model thumbnails and naming in a shared design kit, onboarding patterns for new capabilities, review turnarounds matched to engineering's pace. I ran bug bashes on each release to catch what erodes trust: a missing capability tag, a reference type that shouldn't appear for video, a tooltip clipping at a panel edge.

Growing with the models

The surface absorbed bigger capabilities without changing shape. Duration choices, 4K resolution, frame-rate control for professional workflows, and audio, both native model audio and audio references, as generated video pushed toward finished output.

The outcome

The surface held. The squad formed in April and built video generation from the ground up on an image-first platform, delivering 40+ projects along the way, from Motion 2.0 through Veo 3.1 in 4K, with me as lead designer on every video launch. Each new model landed inside the same system users already knew. What would have been twenty-five redesigns became twenty-five configurations.

The challenge

The most important moment in the product was a dead end. The media viewer is where users see their work at full size and decide what happens next: the most-trafficked surface after generation itself. The production version was a lightbox: the asset, download, organise, delete. Fine for an image platform with a handful of models, not for one spanning images and video across dozens, with editing arriving fast.

Acting on a generation meant closing it and hunting. Upscaling, animating or using an output as a reference meant leaving the view; comparing two outputs meant close and reopen. The prompt, model and settings behind a result weren't visible, so users couldn't learn from what worked or reproduce it. And the replacement had to work identically for an image, a ten-second video, an edited variant, and whatever modality shipped next.

BeforeAfter

My role

Led the UX direction for the redesign:

• Rethought the surface where users view, refine and act on their generations
• Simplified existing patterns while planning for extensibility across modalities, including how editing options appear, adapt and disable per content type
• Aligned design, engineering and product through reviews and prototypes

The process

From lightbox to workspace

A side panel puts prompt, model, settings and metadata beside the asset, with the highest-value actions one click away. Upscale, create video, use as guide. Nothing about the asset is a mystery any more, and nothing users want to do next requires leaving. I iterated the panel's hierarchy through reviews with the founders, down to the weighting of the main action.

A rail for browsing

A thumbnail rail turned the viewer into a browser. Moving between generations became a flick of the scroll wheel or a keystroke instead of close-and-reopen, so reviewing a batch means comparing rather than navigating. Global wheel-to-browse had to coexist with scrolling inside an expanded prompt, and I caught and drove the fix when the two fought each other.

One view, every modality

The layout is defined as rules, not as an image view with video bolted on. A media area that adapts to any aspect ratio and playback type, panel modules that appear per asset (video gets its player and format options, images their guide and upscale paths), and a conversational iteration bar where users describe a change in plain language. Options an asset can't use disable gracefully instead of disappearing, the same principle I later scaled across video generation.

The outcome

Revamped the Focus Modal, which fundamentally changed how users discover the platform and interact with Leo across image and video.

The redesigned viewer shipped as the centre of gravity for the whole platform. Browsing, understanding, editing and acting on generations all happen without leaving. It carried images and video from day one and absorbed the editing capabilities that followed.

The challenge

Professional video needs sound, and the workflow broke at its final step. Video outputs were getting long enough, and good enough, for real work, but a soundtrack meant leaving for another tool. Audio generation closed that loop and opened a new creative surface of its own. It was a true zero-to-one: Leonardo had never shipped a modality without a visual result.

Audio has no thumbnail. Grids, previews and the media viewer all assumed a result you judge at a glance; audio is judged over time, by listening. It needed its own way to be browsed, previewed and compared without fragmenting the product. And one surface had to serve two mental models: generating from scratch (a blank prompt) and steering from a reference track (closer to editing).

My role

Led the design of the audio generation experience:

• Defined how a modality with no visual output is browsed, previewed and compared in an image-first product
• Designed the controls (duration, instrumental mode, quantity) so musicians and non-musicians both get to a good result
• Worked with research, engineering and product to ground the design in what the models could reliably deliver

The process

Designing for something you can't see

The card became the product. A waveform, file name and duration readable at a glance, with play and scrub built in. Audio cards follow the same grid logic as images and videos, so a new track lands exactly where a new image would. Names, durations and waveforms do real identification work when comparing files, down to how duplicate names resolve so two tracks never become indistinguishable.

Two mental models, one surface

Creating from scratch and steering from a reference live in the same flow. The reference is an input you add, and every control keeps its meaning in both modes. One surface to learn, two ways in.

Controls that speak human

The model exposed parameters; creators needed decisions. Raw settings became duration, an instrumental toggle for music without vocals, and how many variations to generate, with defaults that give a first-time user a good result from nothing but a prompt, and seeds for anyone chasing repeatability.

Native, not bolted on

The fastest way to make a new modality fail is to make it feel like a separate app. Audio reused the prompt-first flow, the same generation history and the same system, so anyone who had generated an image already knew how to generate a track.

The outcome

Audio shipped in mid-2026 as Leonardo's first non-visual modality: three models at launch, spanning music, sound effects and dialogue, on one generation surface. Because the design abstracted the surface from any single model, a fourth model slotted in within weeks of launch without a redesign.

Within days of shipping, teams were demoing audio to Leonardo's largest enterprise customers, and it became how creators finish a video with its soundtrack without leaving the product.

The challenge

Asking people everywhere what future they want to help create, and giving a voice and a platform to those goals. It starts with one simple question: what's one goal you'd like to see achieved in your lifetime, for your community, your country, or the world?

Phase 1 was a single landing page. Phase 2 arrived with a hard campaign deadline, a named next region, and a growing dataset nobody was seeing. Three constraints shaped everything: a fixed timeline, future regions certain but undefined, and public, anonymous submissions of any length and language, arriving at an unpredictable rate.

Underneath the brief sat three design problems:

• Scale without rebuilds. Duplicating the site per region multiplies design and engineering cost and lets regions drift apart. Marginal cost per region had to approach zero.
• Content we don't control. One word or a thousand characters, in any language or script, including right-to-left. Localisation was a first-class requirement from day one, and the layout had to absorb all of it while still feeling like it came from real people.
• Data that starts at zero. A region launches with no submissions and grows to thousands. The same page had to be honest and compelling at both ends of that curve.

BeforeAfter

My role

Drove the UX of the platform's hardest surfaces:

• Defined how unpredictable, user-generated content is abstracted into a system the interface can rely on
• Built the design system underneath the platform: colour, type and spacing as variables mapped per region, on a fluid layout, so every new region ships fast and stays consistent
• Shaped the data visualisation: how community themes are ranked, when a pattern is real enough to show, and how the story stays honest at any volume
• Simplified submission to the fewest possible decisions (every field optional, anonymous), on one interaction model that survives translation, right-to-left scripts and every viewport
• Set the design system up so AI agents could build working prototypes with it, and helped drive AI enablement and change management across product design
• Kept design, engineering and marketing aligned so system decisions held under campaign deadlines

The process

A platform, not a page

We weighed a fast bespoke USA page against a themable system, and chose the system. It cost more upfront and made every region after it nearly free. The same foundation shipped the USA, Wollongong and a global edition.

A design system built to scale fast

Colours, type and spacing were defined once as variables and mapped per region, on a fluid layout that holds from mobile to widescreen. That is what made "a region is a configuration" true in practice. Launching a region means setting a handful of values, not redrawing screens, so every region looks like its own while staying consistent underneath. It also kept us fast under a fixed campaign deadline: a change was made once and carried across every region and every viewport.

We stress-tested the claim with the hardest case available: a full right-to-left Arabic localisation. If the system survived RTL, any new region was trivial.

From design system to agentic prototypes

Because the system lives as variables, components and written rules rather than pictures in a file, AI agents can read it as well as people. That made it the foundation for the team's agentic product prototypes: agents build working prototypes on the real tokens and components, so what comes out looks and behaves like the product instead of a generic mock-up.

Getting there was as much change management as design. I helped product designers move from static mock-ups to prototyping with agents, starting from a system they already knew, so trying the new workflow didn't mean giving up the craft standards they cared about.

Content we don't control

Goals render as handwritten postcards. One handwriting font per region, encoded as a component so the decision outlived the design file. Postcard sizing followed the real 1,000-character data, not idealised one-liners.

Data that stays honest

Theme bars scale proportionally, labels relocate when bars get small, and the display caps at four categories. Past that, slivers read as noise. Since a region's launch day is its emptiest day, the zero and low-data states are an invitation to participate, with a submission flow that lowers every barrier: all fields optional, anonymous in seconds.

The outcome

Phase 2 shipped on deadline in mid-2026, and the systems bet paid off. The same variables and fluid layout carry every region, so each new one is a configuration pass rather than a design project, and the phase that followed was built directly on the system rather than rebuilt. The same system now does a second job: it is what the team's agentic product prototypes are built on.

Reflection

When a brief says "relaunch the page," the leverage is usually in questioning the noun. Reframing a page as a platform cost us harder conversations early and saved every team that came after. It also stayed collaborative under pressure, with designers and engineers resolving edge cases together in threads rather than throwing files over a wall. The bigger lesson came later: a design system written clearly enough for agents to use is also the easiest way to bring a team into working with AI.

A bit about me

Outside of product work I paint: my pieces have shown at the Stencil Art Prize in Sydney and at fortyfivedownstairs in Melbourne, and that art practice is where a lot of my visual instincts come from.

The rest of my time goes to travel, cooking, and camping: I'm happiest planning the next trip, feeding people, or somewhere off-grid with a tent.

I also volunteer with DanceWize, supporting safer festival and event communities.

Experience

• Product Designer

  Canva · Sydney

  2026 – Present

  Product designer in Canva's MarTech group, helping drive AI enablement and change management across product design. I build the design systems that power our agentic product prototypes, and help the team change how it designs with AI day to day.

• Product Designer

  Leonardo.Ai (Canva) · Sydney

  2024 – 2026

  Designed across the core generation experience of an AI creative platform that grew to 75M users. Lead designer on video generation (from a single-model modal to a surface that carried 15M+ generations in its first year) and on Leonardo's first audio modality.

• Senior Digital Designer

  Webqem · Sydney

  2023 – 2024

  Led high-fidelity prototypes, wireframes, UI and brand guidelines for digital products at a Sydney digital agency, across clients in advertising, education and Web3, translating complex briefs into clear interfaces on tight timelines.

• Product Designer

  Hello Coach · Sydney

  2023

  Redesigned the member dashboard into the working home of the coaching journey: guided onboarding, coach discovery across eight life areas, status-driven session management and goal tracking on one screen.

• Senior Product Designer

  NFTuence · Remote (UK)

  2022 – 2023

  Designed apps and websites end to end for a Web3 agency, from user flows to high-fidelity prototypes, and led ideation for new NFT products, including pitch work that helped win the MUSESS project.

• Earlier design roles

  Art direction & graphic design

  2007 – 2019

  Art director and graphic designer across brands, agencies and startups, including Oticon, RVD, Digistyle and ZoodFood: the visual and brand foundation behind the product craft.

Education

Bachelor of Graphic Design.

Technical skills

Exhibitions & volunteering

• Stencil Art Prize · Finalist, Sydney, 2019
• SMALL, fortyfivedownstairs · Group exhibition, Melbourne, 2022
• DanceWize · Volunteer