“Job searching doesn't reward quality. It rewards endurance.”
Every application runs the same ritual: read the JD, revise the same CV, write a cover letter that sounds like the last one, fill fifteen form fields manually. Fifty applications. Fifty repetitions.
Existing tools try to fix this and fail. Keyword optimisers make you sound like the job description. AI cover letter generators produce plausible, generic output. Auto-apply bots fire without the candidate knowing what was submitted. The shared flaw: all treat applications as document problems. None treat the candidate as the source of truth.
Careful tailoring takes 1 to 3 hours per application. Multiply that across a real job search.
By application 30, candidates aren't presenting themselves. They're performing a script.
Quality and volume pull against each other. The more you apply, the less each one sounds like you.
A good recruiter doesn't open with your CV. They start with who you are, what you've built, where you're going, then help you present that self in context.
Ply.ai builds on that principle. Before any job is imported, the Candidate Profile exists: career history, skills, projects, goals, voice, Master CV. Every output draws from it. The candidate isn't reacting to the job. The job is evaluated against a complete picture of the person.
A persistent map of the candidate that applications can reference without inventing a new person each time.
Knowledge structure
Every application starts from the same persistent understanding.
Roles
Companies
Scope
Progression
WUZO
WiiTH
Ply
Product Design
Interaction Design
React
AI Workflows
Product Designer
AI SaaS
London
Tone
Working style
Role filters
Experience
Education
Achievements
Job Description
Tailored application
The candidate remains the source of truth.
Every output is composed from the persistent record, not from a generic application pattern.
Four layers, each with a single clear responsibility. Separating concerns meant each layer could be reasoned about, tested, and improved independently.
Applications span hours, days, sometimes weeks. A wizard asks candidates to finish everything in one sitting, then forgets the state between sessions. Ply.ai needed to behave like a workspace because an application is a lifecycle, not a form.
The product got structurally more complex so the candidate's experience could become simpler.
AI surfaces items from the candidate's own record — nothing invented.
AI proposes.
Candidate reviews.
Candidate approves.
Each answer approved independently — Draft, Approved, or Submitted.
The earliest versions exposed too much. Six visible stages. Candidates were managing the system's process instead of their own decisions. The product got simpler as the thinking got clearer.
Earlier model
· Analyse
· Evaluate
· Tailor
· Review
· Export
· Apply
6 stages · Candidate manages the process
Current model
· Opportunity analysed
· Documents ready
· Apply with Ply
Candidate sees outcomes.
Stable fields detected and filled — candidate retains control throughout.
"The hardest part wasn't designing AI. It was deciding where AI should stop."
Every time the product got faster, I had to ask whether a decision had moved away from the candidate without their awareness. Speed isn't always the right optimisation.
Hide complexity badly and candidates see something broken, not something simple.
Building across product, UX, and code meant constraints were real. Technical limits shaped design decisions.
Some steps belong to the system. Some belong to the candidate. The design work is deciding which is which.
The Candidate Profile onboarding. The system's quality ceiling is a design problem before a technical one — it deserved more attention in the earlier stages.
The visual weight of Draft → Approved → Submitted. The states exist. Their significance to the candidate's sense of progress could be communicated more clearly.
These aren't features. They're extensions of one principle: help the candidate show up as themselves, more effectively, every time.
The Candidate Profile improves with every application, building a compounding record across multiple job searches, not just one.
The same person, presented coherently across every application, format, and channel.
Progressively wider browser intelligence, earned only as the system demonstrates reliability.
The same Candidate Profile that builds applications prepares candidates for the conversations that follow.
Graphic & UI/UXDesigned UI/UX for WUZO, a fintech platform featuring peer-to-peer payments, bill splitting, and QR code payments. Developed site maps, flowcharts for transaction history and KYC process, and built scalable design systems for consistency across platforms.
Product DesignDirected comprehensive design for WiiTH, an app linking offline experiences with real-life connections. Created user flows, wireframes, and high-fidelity UI. Established and maintained a design system to enhance visual consistency and expedite development.
Brand IdentityCreated the full brand identity for PHYASE //, a Hong Kong physio and sport performance brand — from concept naming, tagline development (EN & Traditional Chinese), to logo construction, photography direction, and brand guidelines delivery.
