A concept and front-end design exploration for Wealthsimple's AI Builder program: how AI-assisted behavioural analysis could live inside Wealthsimple's existing product without adding friction, compromising user autonomy, or requiring any new data collection.
Scope, up front: this is a front-end flow built on mocked data — not a working product. I designed and built the experience (onboarding, the in-app analytics and check-in views, the notification moment, the budget and advisor screens) to show how the idea would look, feel, and flow end to end. There is no live backend, model pipeline, or real account data behind it. The value here is the idea and the direction, demonstrated concretely.
Most fintech AI is built around prediction or prescription — tell users what to do, when to do it, and why they're wrong. This takes a different approach.
The premise: Wealthsimple already captures everything it would need — transaction data, portfolio actions, savings patterns, account activity. The gap isn't data, it's interpretation. This concept sits between the data and the user, surfaces patterns they might not have noticed, and asks questions rather than giving orders.
The result is a check-in that feels like a conversation with someone who's been paying attention — not a compliance warning or a sales pitch. It notices when a person's own financial behaviour drifts (spending creeping up, savings slipping off-track, reactive portfolio moves) and reflects that back, plainly.
- Onboarding — goal → time horizon → risk tolerance → investment experience → profile review; optional personalisation (life stage, financial stress, check-in frequency).
- Budget setup — a 50/30/20 framework scaffolded to the user's income, fully editable.
- Home — a quiet dashboard; a notification card triggers the 90-day check-in only when drift is detected. No noise when nothing needs attention.
- Analysis view — savings actual-vs-on-track, category-level spending shifts, a portfolio activity log, and a plain-language AI summary in three parts: what we noticed, what it might mean, one thing to consider. A soft advisor prompt is available but never pushed.
- Budget screen — live allocation editing, lockable categories (so suggestions never touch what the user protects), and paginated pattern detection.
- Advisor view — an internal-only screen with a behavioural-signal breakdown, a confidence score, and risk flags, framed around client-initiated contact only.
Design principles throughout: no new data collection, the user stays in control, and intervention is opt-in, not imposed.
- App / UI: Streamlit + a custom CSS design system (Wealthsimple dark theme)
- Plain-language summaries: Claude (Anthropic) via API, run over mocked behavioural signals
- Charts: Plotly
All transactions, signals, and patterns are mocked and intentionally realistic (a recurring purchase, home-maintenance spend miscategorised as shopping, a new subscription appearing two months running). In production these would come from Wealthsimple's existing pipeline — no new collection required.
The same loop should generalise from people to AI agents. Swap dollars for tokens and compute: what is an agent actually spending its budget on, and is that what it was asked to do? How far does it drift from its original goal across a long task? Same machinery — establish a baseline, measure divergence, surface it legibly, keep a human in the loop — applied as an oversight signal for agentic systems. That's the direction I'm most interested in carrying forward.
MIT