Flourish

Inspiration

It's 11:47 PM. She just got her first paycheck, opens her brokerage app, and stares at a stock ticker she bought because it was the only company name she'd ever heard men at work talk about. That moment was our starting point. Fidelity's Smart Money Moves challenge asked us to help early-career women build smart money habits, connect finances to career growth, build community, and feel confident making their first money moves, and we kept coming back to one insight: most people don't fail at investing because it's hard, they fail because it feels intimidating. Confusion, not incapability, is the real barrier. So instead of building another investing app, we built the layer that removes the intimidation from the one she already has.

What it does

Flourish is an AI confidence layer that sits on top of a real Fidelity account. It has four pillars:

-Portfolio Hub: connects to a Fidelity account and explains it in plain English first, a daily return decoded ("what's a return?"), a correlation insight like "your stocks are close friends, and that's not helping you," and a Flourish forecast, with the full charts and correlation matrix available one tap away for anyone who wants to see the data behind it.

-Discover: a "women invest in women" hub surfacing women-led companies, female CEOs, and women-focused funds (including Fidelity's own FWOMX), fit-checked against the user's actual holdings so discovery turns into a next step instead of a dead end.

-Translate: an AI news translator that turns confusing headlines (SpaceX IPO, Fed rate decisions) into plain English at three reading levels, plus a Chrome extension concept that brings the same simplification to any page on the web.

-Coach: a RAG-grounded chatbot with a browsable knowledge base, a Quick Prep tool that generates a pocket cheat-sheet before any money conversation, salary negotiation rehearsal, and an After Dark mode that softens the tone for the questions users are embarrassed to ask out loud.

How we built it

-Frontend: React 18 + Vite + Tailwind CSS, with Recharts for the portfolio visualizations.

-Backend: a zero-dependency Java 21 HTTP server we wrote from scratch. It proxies every AI request to the Anthropic API so the key never touches the browser, serves mock client account data, and persists saved interests, all through a small REST API.

-AI: live Claude (claude-sonnet-4-6) powers every AI surface, translation, coaching, portfolio health reads, and Quick Prep. Every system prompt is grounded with the user's profile, linked portfolio, and saved interests, and constrained to education only, never a buy or sell recommendation.

-Data: a JSON-based mock data layer standing in for a production database, structured the way a real Supabase/Postgres schema would be, so client data flows through a real API instead of being hardcoded into the UI.

Challenges we ran into

Keeping the platform from becoming overwhelming was the hardest design problem, we started with a much denser dashboard and kept stripping it back to "insight first, data on request," since our target user is intimidated by finance apps, not looking for more data density. On the technical side, moving our AI calls from a direct browser-to-Anthropic connection to a real backend mid-build meant rearchitecting our data flow without breaking the UI, a good tradeoff for a much more honest security story. We also had to be deliberate about compliance: every AI prompt had to be worded to give education and scenario exploration, never advice, which matters a lot when you're prototyping something that could sit next to a real brokerage account.

Accomplishments that we're proud of

-A fully live AI layer, not a mockup, running through our own backend.

-A real security-conscious architecture: the API key lives server-side, client IDs are sanitized against path traversal, and the whole system is designed around OAuth-style delegated access rather than ever touching a real credential.

-A cohesive personalization loop: a company hearted in Discover is fit-checked against the real portfolio, flows into a "how do I go from here" coach conversation, and feeds a visible confidence meter, four features working as one system instead of four disconnected tabs.

-After Dark mode, which we think is the most honest feature on the platform: the same coach, a softer tone, for the 11 PM questions nobody asks in daylight.

What we learned

We learned that the hardest part of building for financial confidence isn't the AI, it's the restraint: every feature we cut (stock predictions, a poker-themed "decode Wall Street" game) made the product stronger by keeping it educational and compliant rather than flashy. We also learned a lot about the real architecture behind fintech account linking, OAuth delegation, scoped tokens, row-level data isolation, and designed our backend to that standard even though our 24-hour build necessarily uses mock data instead of real Fidelity credentials.

What's next for Flourish

-Real account integration via Plaid or Fidelity's own OAuth flow, replacing our mock data layer with live, read-only, revocable access.

-A production RAG pipeline: embedding our curated knowledge base into a vector store instead of the flat JSON corpus we used for the prototype.

-Shipping the Chrome extension concept as a real browser extension.

-Investment Circles: small, women-first communities (Women at [School], First Job, Grad School) built on the same anonymous, judgment-free model as After Dark.

-A compliance review with Fidelity's legal team on every AI system prompt before any of this touches a real customer.