Selected Work
Built with judgment, shipped with AI.
Building Spendflo's Agent Builder
Designed the Agent Builder inside Flo AI: how a team describes an agent, configures its instructions and limits, tests it, deploys it and traces every run, reusing Spendflo's existing product patterns end to end.
Spendflo Demo Environment
Turned Spendflo's existing Figma designs into a reliable, scenario-driven demo environment using Claude Code, cutting Sales' dependency on Dev and an unstable shared demo setup.
Redesign Navigation
Spendflo's top navigation hit a ceiling as features multiplied within each module. Audited, tested, and validated a left-nav information architecture that cut rage clicks 45% and quick backs 86%.
Rebuilding Spendflo's Library
Spendflo's Design Language System had drifted: duplicated components, inconsistent spacing, and no documented usage patterns. Led a full DLS revamp with a 2-person design team, cutting UI-related bug reports by 65%.
Experience
Where the craft got tested.
Spendflo
Lead design for B2B procurement: AI features, workflows, and the design system.
- Built a self-service demo environment with Claude Code, so Sales could run demos without waiting on Dev
- Rebuilt the design system from the ground up: 65% fewer UI bugs, 30% faster design execution
- Led AI workflows and conversational UX design, including agent experiences
- Led the navigation redesign, moving Spendflo from a top nav to a left nav so the platform could scale
- Own design for invoices, bills and payments
IceApple Technology Solutions
Designed end-to-end UX for AI chatbot and search experiences within a real estate platform, simplifying complex property discovery workflows.
- Owned conversational interface design: chatbot flows, intent mapping, fallback states, error recovery
- Partnered with engineering to deliver responsive, user-centred experiences across multiple product lines
- Contributed to branding and visual identity for JJ Foods and MITRA AI
How I Work
Lessons from building with AI.
A few things that only became clear after actually shipping real products with AI in the loop.
Generation is cheap, judgment isn't.
AI can produce a dozen directions in seconds. Picking the right one, and knowing when to override it, is still the job.
The fastest path is prototyping in code.
Building the Spendflo demo environment, and this site, directly with Claude Code closed the gap between what's designed and what ships.
Documentation is leverage, not overhead.
Centralizing usage guidelines and auto layout in the DLS revamp sped up the whole design team, not just the AI.
AI changes the interaction model, not just the tooling.
Conversational and agent-driven interfaces need fallback states and trust-building that static screens never did.
My Design Process