Vested

The Problem

The framing for most personal finance products is reactive: you open the app, you see your balance, nothing happens. The more sophisticated ones send a push notification after something has already occurred. Vested was built around a different premise — that the most valuable moment in personal finance is before the decision, not after. What would it look like if your financial tools watched for opportunities and surfaced them without waiting to be asked? And what would it mean to do that responsibly? A financial AI that hedges everything is useless. One that overpromises is dangerous. The interesting design space is in the middle: a system that's genuinely helpful precisely because it knows its own limits.

Why This Needed AI

There are things this product does that simply weren't possible — or were prohibitively expensive — before large language models. Three in particular shaped the design from the ground up.

1. 01Proactive pattern recognition: The system watches across multiple accounts — a dip in 401(k) contributions, an approaching PLESA limit, a market move that touches a specific holding — and surfaces it as a prompt. No user query required. This isn't a notification; it's the product doing its job.
2. 02Personalized math at the decision point: When a user asks 'what if I retire at 60?' they shouldn't get a generic article. They should get a calculation built from their actual accounts, contribution rate, and timeline — surfaced right now, in the conversation flow. That's only possible when the model has context.
3. 03Calibrated confidence: The AI says 'I'm not sure' when it is, and routes to a human CFP when the question warrants one. This isn't a failure state — it's a design decision. Making the system's uncertainty visible is what makes it trustworthy.

Try It

The prototype is fully interactive — start from the dashboard, try the AI chat, ask about your accounts, or type 'talk to a person' to see the human handoff.

What This Prototype Doesn't Do

This is a design prototype, not a product. Honest accounting of scope before you read anything else into it.

• Connect to real accounts — all data is simulated
• Navigate a regulatory environment — FINRA, SEC, and state regulations would significantly constrain much of this interaction design
• Build trust over time — the prototype starts with a pre-populated history, but a real product would earn every one of those sessions
• Run on a real model — AI responses are canned, not generated; the prototype demonstrates the interaction pattern, not the underlying intelligence

Key Design Decisions

Five decisions shaped how the prototype feels — and each one was a direct answer to the trust question.

What I Learned

The hardest part of this project wasn't the prototype — it was deciding what the AI should and shouldn't do. Every design decision eventually came back to the same question: is this building trust, or performing it? The scan animation, the confidence color grammar, the two-beat advisor handoff — all of them were answers to that question in different registers. I also learned that the human handoff isn't a fallback. It's a product feature. Designing it with care — giving the advisor context, making the transition visible, not pretending it isn't happening — turned what could feel like a failure state into a moment that actually strengthens confidence in the system. Fintech is a domain where the cost of getting UX wrong is real. That constraint made every decision more interesting.