Conducted 12 moderated user interviews with creators across verticals (education, fitness, business coaching). Reviewed 3 months of Hotjar recordings to identify rage-click hotspots and drop-off patterns. Analysed support ticket themes with the CX team to create a quantified problem stack.

Card-sorted the full feature set with 20 existing creators to identify their mental models. Restructured the IA around three core jobs-to-be-done: Create, Manage, and Grow. Reduced primary navigation depth from 4 levels to 2.

Built a component library from scratch in Figma — covering 60+ components with variants, states, and accessibility annotations. Established design tokens (colour, spacing, typography) mapped to CSS variables for dev handoff in Figma Dev Mode.

Built interactive high-fidelity prototypes for the 5 highest-impact flows. Ran two rounds of unmoderated usability tests (n=30 per round via Maze) and iterated on findings before dev handoff. Measured task completion rate from 58% to 91% on core flows.

Shipped in 6 two-week sprint cycles alongside two engineers. Used feature flags to roll out changes progressively, monitoring engagement and error rates per feature. Maintained a living design spec with inline Loom annotations for complex interactions.

The original homepage opened with a generic tagline about "event insights." I restructured the hero to lead with the AI-specific outcome — what changes for the buyer after using Snapsight. The new headline tested significantly better in copy testing: clarity score up from 4.1 to 8.6/10 with target users.

The original design used a dated blue-and-white palette with stock photography. I developed a new visual system: dark-mode primary with a signature accent, custom data visualisation components as decorative elements, and a motion language that implied intelligence without being gimmicky.

Rewrote all body copy, CTA labels, and feature descriptions to focus on buyer outcomes rather than product capabilities. Replaced "AI-powered sentiment engine" with "Know exactly how your audience felt — instantly." Every CTA now states the next concrete step, not an abstract invitation.

Built a Figma component library the marketing team could extend without breaking the visual system. Created a Framer CMS structure so blog posts, case studies, and feature pages could be added by non-designers — reducing design dependency for content updates.

Co-authored product requirements documents with the founders, defining the MVP feature set by ruthlessly cutting anything that wasn't essential for conference validation. Mapped user flows for 3 core personas: Impact Officer, Executive Sponsor, and External Auditor.

Built the design system (tokens, components, patterns) before any screen design began. This allowed parallel workstreams — design could proceed on multiple screens simultaneously without creating inconsistency, and dev could start building components before full flows were designed.

Designed all screens at full fidelity — no wireframe phase given the time constraint. Used Figma's Dev Mode with inline annotations for every interactive state, edge case, and responsive breakpoint. Ran daily standups with the dev team to unblock implementation questions in real time.

Led QA reviews in the final two weeks, logging and prioritising visual bugs and interaction issues. Separated must-fix from nice-to-fix to protect the deadline while ensuring the demo flow was flawless.

Separated create, import, connect, and watch-only journeys at the first decision point. A short PIN setup and explicit blockchain selection give users control without forcing them to understand the complete technical model upfront.

Designed a 3-step send flow with an address confirmation screen showing a visual checksum and ENS name resolution. Added an optional "test transaction" prompt for first-time sends. Clear fee display in fiat equivalent before, not after, confirmation — eliminating the surprise gas fee problem.

Combined buy, swap, cross-chain, and dApp discovery under one trade destination. Chain, provider, token, quantity, slippage, and estimated value follow a clear top-to-bottom sequence so users can understand what will happen before committing.

All interactive elements meet 44×44pt minimum touch target requirements. Bottom-sheet patterns for action confirmation keep critical actions in thumb reach. Bottom navigation optimised for one-handed use — the most common posture for mobile financial apps.

Developed a brand identity that communicated AI capability without feeling intimidating. Logo, colour system, typography, and iconography all built around a "knowledgeable but friendly" positioning — distinguishing Answerflow from the stark, technical aesthetic common in AI tooling.

Replaced a configuration panel metaphor with a step-by-step wizard that guided users through chatbot creation as a narrative: "Give your chatbot a name → Add your content sources → Test it → Deploy." Progress state was always visible. Users never saw a blank canvas.

Designed a live-preview customisation interface — users could change colours, fonts, avatar, and greeting text and see the widget update in real time on the right side of the screen. No guesswork, no "save and check" loop.

Designed the chat widget UI to be visually neutral by default and fully customisable through the dashboard. The widget needed to feel at home on a luxury spa website and a B2B SaaS platform simultaneously — achieved through a themeable token system rather than opinionated defaults.

I used the Antler MENA cohort environment to sharpen the audience, workflow problem, product promise, and MVP scope before committing to interface directions.

I defined the information architecture, intent-based inbox model, interaction patterns, visual language, and reusable components as one coherent product system.

Using AI coding agents, I converted design intent into a functional React application, directing implementation through structured prompts, product specifications, acceptance criteria, and continuous visual QA.

I reviewed generated implementation against the intended user flows, corrected interaction and hierarchy gaps, refined the experience across breakpoints, and deployed the product and marketing surfaces.

Instead of forcing users into a traditional unread/starred/archive workflow, I designed the inbox around real-world mental buckets: Important, Job Applications, Payments, Newsletters, Updates, and Other. This gives structure immediately and reduces the "where do I even start?" problem.

Quick actions like remind, complete, and delete live where the user is already looking. The goal was to remove toolbar hunting and make triage feel lightweight enough to do continuously instead of in long cleanup sessions.

The AI assistant was designed as an on-demand layer over the inbox, not a takeover. Users can search messages, find tasks, draft replies, and summarise threads inside a familiar chat surface while still keeping the original email context visible.

The visual direction uses soft gradients, light surfaces, and simple iconography to make a productivity tool feel approachable. The website and product were designed to tell the same story: this is serious software, but it should still feel easy to start using.

Started with a service blueprint that followed one payment from checkout to vendor settlement. Mapped system events, ownership changes, gateway responses, fee calculation, refunds, and operator interventions.

Aligned customer-facing language, vendor status, admin status, and backend events around one state matrix: initiated, pending, paid, failed, refunded, blocked, and settled.

Built low-fidelity wireflows for customer checkout, vendor earnings, and Superpage operations. Keeping the work intentionally rough helped the team debate logic and hierarchy before visual polish.

Worked through gateway timeouts, fallback routing, incomplete vendor setup, partial refunds, payout holds, and reconciliation mismatches. Each state had to explain what happened and what could happen next.

Reviewed wireflows against gateway behaviour and backend events, resolved ambiguous states, and annotated the expected data and actions for each surface before handoff.