Senior UX consultant with seven years on an enterprise compliance platform. I turn dense domain rules into calm, usable products, design AI features people can check and override, and hand developers designs they can build.
Four projects, each written as a design record: the problem, the process, and the decisions that shaped the result.
An assistant that reviews uploaded evidence before an assessor does, shows where it found each gap, and leaves the final call to a person.
Read case study →
A six-step workflow where AI parses firewall configurations from four vendors, proposes the scope, and turns risky rules into findings people can verify and act on.
Read case study →
Making many-to-many regulatory mapping understandable for client admins and assessors.
Read case study →
Bringing scattered files, letter templates, deadlines and billing into one calm workspace.
Read case study →Clients upload policies and records as evidence for each audit question. Assessors reject weak evidence days later, and the client starts again. I designed an assistant that checks evidence on upload and explains exactly what is missing.
Each audit question comes with specific requirements, such as a minimum password length or a rotation period. Clients upload documents they believe cover it. If something is missing, they find out only when an assessor reviews it, often well after the upload.
An AI check at upload time could close that loop in seconds. The design challenge, before it went into development, was trust: clients had to understand why the AI flagged something, and assessors had to stay in charge of pass or fail.
Wrote out the paths before drawing anything: evidence passes; evidence fails every requirement; the client fixes part of it and re-uploads one file; the evidence doesn't mention a requirement at all.
An introduction screen with an explicit agreement to the AI terms, then a per-question switch, so contributors decide where the assistant runs.
Designed the pre-check panel: in progress, success, and gaps, each gap paired with the requirement it misses and a reference to the exact page.
Re-uploading one file triggers a new check, and the gap count drops as requirements are met. Earlier files and earlier runs stay visible by date.
Assessors see the same pre-check results, with and without gaps, as input to their own review, not as a verdict.
Prototype 5 established the core idea that shipped: check on upload, explain each finding with a reference, let the client fix it, keep the assessor in charge. The product has been refined in several rounds since.
Developed from this prototype and used by clients during evidence collection.
Show how sure the AI is on each finding, and let users flag one as wrong to improve the model.
Placeholder: a measured effect since launch, such as fewer rejected uploads or faster reviews.
Reviewing firewall rules for an audit means reading thousands of rules, written differently by each vendor, and tying every risky one to a compliance control. I designed a workflow where AI does the reading and people stay in control of scope and findings.
The six-step workflow is live, from upload to reports.
Let reviewers mark a finding as a false positive or an accepted risk, and show how confident the AI is in each finding and in the detected scope.
Placeholder: a measured effect, such as review time per firewall before and after.
Clients prepare for audits against several regulations at once. The work is to show how their own controls connect to those regulations without making them learn the data model.
Every regulation brings hundreds of requirements. Every framework a client sets up holds around a hundred controls. One control can satisfy requirements from several regulations, and one requirement can touch several controls.
The system could already do the mapping. The people using it could not see it: admins didn't know which frameworks were ready, and assessors had to leave their work to check how a control was mapped.
Worked through regulations, requirements, controls, frameworks and audits with product leads until I could explain the lifecycle back: a framework is the setup, an audit is where evidence gets collected and assessed.
The client admin lands on the Frameworks list. That made the list the most valuable screen in the feature, more than any new page.
Designed a standalone overview dashboard and a richer list view side by side, then put both in front of the VP. The review shaped the direction (see DEC-01).
Restarted the light version from my own guidelines: framework views, create-framework modals, controls screens and a control detail drawer, all built from one component library.
Added micro-interactions in a separate file. My first pass felt heavy, so I cut it back to quick, quiet feedback on selection, save and state change.
Framework health is visible on the screen admins already land on.
Control-level mapping and evidence live in one drawer.
Placeholder: a measured result, such as support tickets, task time or adoption, once it is available.
The firm's files were spread across offices and folders. Letter templates existed in many versions, and no one shared a calendar. The owner isn't technical, so the system has to feel obvious from day one.
The firm's biggest pain was years of legacy material: templates, document reviews and case files with no single home. On top of that, deadlines lived in separate calendars across four offices.
The goal was a system staff would actually use. It had to start with their most frequent task and grow from there.
Mapped how a matter moves through the office, from intake to letter to invoice, and where things get lost.
Letters on the firm's letterhead were the daily, repeated task, so that became the first module: write, preview, save a PDF draft, send by email.
Designed the system, then a second version in a different visual style for the same brand, so the client could react to real options. The dashboard is being redesigned in Figma next.
Built and tested the app myself on the firm's existing hosting, with documents moving to cloud storage at launch. Kept infrastructure to what the firm can run.
Next step is working alongside the team in Mauritius to import legacy files, train staff and adjust the system to how they actually work.
Letters, documents, deadlines and billing in a single system across four offices.
Started with a familiar daily task to earn trust before adding anything else.
Placeholder: results after go-live and the on-site month, such as files migrated or time saved.
The same five steps on enterprise platforms and small apps. Only the depth changes.
Learn the domain and its rules until I can explain them simply. Complex products fail on the model before they fail on the UI.
Find who logs in first and what they need on that first screen. Choose the smallest valuable scope.
Show real options to stakeholders, then write down the decision and why. Each decision gets an ID so the team can refer back to it.
Work inside a component library with variables, and keep shared patterns identical. Each significant change gets a new version, with the previous one kept intact.
Hand off specs developers can build from. Review the build, then refine interactions until they feel light.
I start projects from my own Figma framework: standard project pages (overview, problem statement, brief) and a branded component library.
I use Claude to explore, prototype and build working software, and hand designs to developers who build with AI coding tools. I'm training toward agentic UX: designing products where AI does part of the work.
Every screen comes with its states, edge cases and the reasoning behind it, so developers don't have to guess.
I'm a senior UX and product designer based in Milton, Ontario. For more than a decade I've designed and built web and mobile products, most recently enterprise compliance software, where the rules are complicated and clarity matters.
I came to design with a front-end background, which is why I care about what happens after handoff. These days I spend much of my time on how AI changes the way products are designed and built, and I put it into practice on real projects.