← All work

Product & Systems Strategy · 2023–Present

Making a Complex System Feel Like One Conversation

AI-assisted product & systems strategy for an identity-rich platform

  • Product thinking
  • AI-assisted design
  • Onboarding
  • Systems translation
  • Technical collaboration
My role
Product/experience lead: framing the problem, mapping the end-to-end journey, defining the staged onboarding and its states, and pairing with engineers to turn concepts into buildable flows.
Audience
A broad, non-expert audience arriving with mixed confidence and device contexts — plus the technical collaborators who build and maintain the flows. Two audiences, one experience.
complex systemone guided sequence
Abstract representation — no proprietary UI shown

Context

As founder of an AI-powered venture (later supported by NSF funding), I owned the experience strategy for a web platform whose value depends on a genuinely complex underlying model — many interacting inputs that combine into a personalized result. The concepts are powerful but easy to overwhelm a first-time user with.

Problem

The system’s depth was its differentiator and its biggest usability risk. A literal, all-at-once interface would expose too many variables and stall people at the exact moment they should feel momentum. The design problem was sequencing, not decoration: how do you reveal a complex model progressively so the first session feels like one guided conversation rather than a form to survive?

Constraints

  • The underlying scoring/logic is proprietary — the UX had to communicate value without exposing the model.
  • Backend intelligence was evolving in parallel, so parts of the experience had to be designed against stubbed/expected data.
  • Small team, fast cadence — decisions had to be legible enough for engineers to build directly from them.
  • Accessibility and clarity were non-negotiable for a diverse first-time audience.

Approach

  • Mapped the full journey first, then aggressively sequenced it — separating "must decide now" from "can resume later" so the required path stays short.
  • Designed the complexity as progressive disclosure: each step asks for one thing, explains why it matters, and shows the payoff before asking for more.
  • Modeled explicit states (empty, pending, returning) so the experience holds up before the backend is fully live.
  • Wrote the flows as clear, reviewable specifications engineers could implement without guessing intent.

Key decisions

  • Split identity/appearance choices from deeper "interview"-style depth so users get a quick win first and depth stays optional and resumable.
  • Represented the still-evolving intelligence with honest pending states instead of fake precision, so the UI never over-promises.
  • Kept a single accent/identity signal consistent across the whole journey so a complex product still reads as one coherent world.

Use of AI

I used AI-assisted workflows throughout — rapid ideation, drafting interaction copy, exploring alternative flows, and modeling content — while holding a hard quality bar on usability, accuracy, and ethics. AI accelerated exploration; judgment about what shipped stayed human.

Accessibility & ethics

Designed for a diverse first-time audience: staged cognitive load, plain-language step framing, honest pending/empty states, and interface controls for text scaling, reduced motion, and higher contrast so the experience adapts to the person rather than the reverse.

Outcome

  • A short required path with resumable depth — the complex model reaches the user as a guided sequence, not a wall of inputs.
  • Specifications concrete enough that engineering could build directly from them, reducing back-and-forth.
  • A pattern language (staged reveal, consistent identity signal, honest states) reused across multiple product surfaces. [add metric: e.g., completion or time-to-first-value once measured]

Reflection

The hardest part of a complex product isn’t adding features — it’s deciding what a person is allowed to ignore, and when. Designing the sequence and the "not yet" states was where most of the usability actually came from.