I’m Zean Lai, a UX & strategic design generalist with a toolkit that spans research, prototyping and data-informed storytelling. I turn fuzzy problems into shippable, ethical solutions—lately across social media help-seeking, audio/DSP experiments, and accessibility-first game concepts. I care about clear problem framing, measurable outcomes, and craft that holds up in the real world.
This project began with the gap between emotional support and actionable guidance in online help-seeking. In a small team I led the research frame and taxonomy—separating substantive help, empathy, vague, and irrelevant responses—then translated insights into a lightweight browser-side prototype that nudges commenters toward concrete steps and credible resources. My core contributions were: sampling strategy, labeling rules, and the interaction model for “nudge to substance.” The hardest trade-off was balancing measurable uplift (more actionable replies) with ethical safeguards (tone, safety, consent). We iterated on prompt wording, source quality checks, and an opt-in flow for creators. I learned to argue for evidence without erasing empathy, and to communicate constraints clearly to non-technical stakeholders.
I built a “robot voice” processor in MATLAB to explore real-time DSP under tight latency. The chain combines band-vocoding (8–12 analysis bands) with a fixed-pitch carrier, a low-rate ring-mod layer (≈30–40 Hz) for metallic tremor, and optional bit-depth reduction. My design goal was to keep latency under ~20–30 ms while preserving intelligibility. Early versions used naïve resampling and produced harsh artifacts; I switched to analysis/synthesis with windowed OLA, tuned window/hop, and added envelope smoothing to stabilize consonants. Spectrogram checks and quick A/B listening guided each tweak. The biggest lesson was how small parameter moves—band count, envelope attack/decay, carrier type—swing between “cool robot” and “unintelligible buzz.” If I revisit this, I’d add formant-preserving pitch handling and port the chain to a real-time framework (e.g., Audio Toolbox objects or JUCE) with MIDI-mappable controls for performance.
I helped shape a small furniture brand’s digital voice and a repeatable content system. Deliverables included a bilingual caption library, reusable hashtag sets, and a lightweight photo style guide so collaborators could ship consistent posts. I also tracked simple engagement signals (saves/DMs) to refine hooks and calls-to-action. The hard part was balancing “Hamptons/Coastal” aesthetics with operational constraints—stock, lead times, and seasonal photos. I learned to design content as a system: templates, defaults, and feedback loops that hold quality under real-world pressure.
Before Dusk is an unfinished game concept exploring Alzheimer’s through mechanics rather than exposition. The core loop links fading memory to dusk: as light falls, environmental cues (names, objects, wayfinding marks) gradually blur or swap, forcing players to rely on routine, sound, and color rather than short-term recall. My intent was to model cognitive load ethically—not to “simulate” the illness, but to invite empathy for uncertainty and dependence on caregivers. I built low-fi prototypes of three scenes (kitchen, hallway, garden) and tested input simplicity (two-button scheme) plus accessibility (high-contrast mode, audio cues). The hardest trade-off was balancing challenge with dignity—avoiding frustration loops and stereotyped portrayals. I learned to set guardrails early (content warnings, opt-out paths, reflective endings) and to define “success” as emotional clarity, not difficulty spikes. Next steps: paper-playtests with proxy users, refine cue hierarchy, and document an ethics checklist with sign-off points.