Helios So — Product Designer × Engineer
Who am I
I design it. I build it. AI accelerates it.
Product designer from Hong Kong, based in London. In UI/UX since 2016 — eight years across fintech and banking, owning end-to-end design on a regional mobile-banking app (2M+ users), a digital wealth-management platform, and an intent-aware help portal for a financial-services group. Currently designing in the neurodiversity space, on a mental-health product.
Two deliberate moves to keep the work honest. I believe in data-driven design — so I'm a year into a Master of Data Science at CU Boulder. I believe a designer needs technical fluency — so I shipped through the Northcoders full-stack bootcamp in 2025.
I frame the problem in a spec, map the technical constraints, and choose the design that fits the stack — not just the screen. I design; AI coders close the loop to working code.
A privacy-first local-LLM chatbot (GitHub ↗) and an AI workbench that orchestrates Claude Code and Codex as a two-head agent (GitHub ↗). I partner closely with engineers and harness AI as a force multiplier to ship high-quality work faster, end to end.
I care about shipping, about systems, and about cross-cultural collaboration. Looking for a role where both halves — designer and engineer — run at once.
What I offer
Building blocks for the team — PO to dev.
I'm a UI/UX designer. What I deliver is every building block the team needs to ship — a spec the PO can test, prototypes the stakeholders can react to, and a design system engineering doesn't have to second-guess. AI accelerates each step; data closes the loop.
Write the spec, before drawing screens. Problem framing, scope, success metrics, roadmap shape. PO and stakeholders get a brief that's testable, not just a deck.
Ten years across fintech, banking, and now the neurodiversity space — owning user research, IA, flows, interaction, and prototyping end to end. Accessible and responsive by default.
Design systems and dev-ready handoff. Figma libraries bound to code tokens, documented for engineers — not just designers. Because I've written frontend code (React, native, TypeScript, Flutter, Electron), my files are ones developers can build, not interpret.
AI as a force multiplier across every artefact above — spec drafts, prototype iterations, review passes, handoff polish. Process matters more than tools.
A year into a Master of Data Science at CU Boulder. I bring a data lens to design — telemetry, A/B tests, funnel analysis — so the next iteration is informed by evidence, not opinion.
One designer, every building block. The team ships faster because nothing gets lost in the handoff.
My workflow
A small loop, run many times.
I don't believe in one-shot design. I believe in a short loop you can run in a day if the stakes are low, and in a week if they're not. Same five steps every lap — research, explore, decide, build, ship — but each lap puts different solutions on the table and picks the direction, so the work compounds instead of drifting.
Interviews, stakeholder maps, research synthesis. Write the problem before drawing anything — if I can't say it in one sentence, the loop hasn't started yet.
Multiple solutions on the table — every time. Divergent sketches, AI-assisted prototypes, option trees. Generate ten; keep three. Different shapes, not three colour variants of the same thing.
Pick the direction — every time. Every lap exits with one option chosen and the trade-offs explicit in writing. Scope the MVP, name what's out, commit to the spec. Decisiveness is the habit, not the moment.
Design system first, screens second, engineering in parallel — with PO and dev close enough to react in the same week. Files are dev-ready, not "designer-ready."
Telemetry, feedback, usability sessions. The loop never ends; it just gets tighter. What we learn becomes the spec for the next lap.
The loop is older than AI. What AI changed is the cost of each lap — so I run more of them.
Why work w/ me
Why work with me.
Not a translator between design and engineering — I'm on both sides of the handoff, so there's no handoff. The design survives the translation layer because I'm the one writing it.
The 3× comes from how I run the loop with AI, not from naming five tools. Process matters more than apps.
Built in Hong Kong, learning in London, studying in the US. Comfortable in ambiguous, cross-border teams — and used to translating between them.