Finguard AI

Inspiration

One of our team members grew up watching their parents make major financial decisions — switching jobs, taking loans, moving cities — with basically no information. Not because they were irresponsible. Because every financial tool they found assumed you already knew the vocabulary, already had a financial advisor, and already understood what an emergency fund or an umbrella policy even was. That's not a niche problem. 57% of Americans can't cover a $1,000 emergency from savings. 40 million renters don't have renters insurance. Most people making the biggest financial decisions of their lives are doing it alone, on Google, at midnight. We wanted to build the tool that the person deserved. Not a budgeting spreadsheet. Not another fintech dashboard full of jargon. A coach — one that speaks plainly, meets people where they are, and gives them a next step they can actually take.

What it does

FinGuard AI is a free, AI-powered financial wellness coach built for people who have never had access to one. In under 3 minutes, it does three things:

Financial Health Assessment You answer a short quiz: age, income, housing situation, zip code, and how much you have saved. FinGuard scores you across four dimensions: emergency fund readiness, insurance coverage, disaster risk (based on your actual zip code), and overall financial health. No jargon. No judgment.
90-Day Action Plan Instead of leaving you with a score, FinGuard generates a personalized week-by-week action plan with specific, time-boxed tasks like "open a high-yield savings account this week" or "get a renters insurance quote, average $15/month."
Life Impact Simulator This is the feature we're most proud of. Before you buy a car, move to a new city, take a loan, or change jobs, you can simulate the decision across three parallel futures: Safe, Balanced, and Risky. The simulator calculates how each path affects your finances, stress level, risk exposure, and life goals. It also runs a shock test — what happens if you lose your job mid-decision? So you can see the breakpoint before you commit. Throughout everything, the AI Coach (powered by Google Gemini) is available to answer questions in plain English, with voice responses through ElevenLabs for accessibility.

How we built it

Frontend: React + Vite, deployed on Vercel. We used Recharts for the dashboard visualizations — score rings, radar charts, bar charts, and area charts for timeline simulation. Backend: Python FastAPI, deployed on Render. We built two separate engines, the FinGuard risk scoring engine (emergency fund, insurance gap analysis, flood risk by ZIP code) and the Life Impact Simulator (a deterministic decision model that calculates financial, stress, risk, goal, and resilience scores across three scenario modifiers). AI: Google Gemini via direct REST API for the AI Coach, with a financial wellness system prompt tuned for Grade 8 reading level responses. We built a robust fallback system covering 9 keyword categories, so the coach always gives a useful answer even when the API quota runs out. Voice: ElevenLabs text-to-speech on AI Coach responses for accessibility. Auth: Auth0 for authentication — login is optional for the assessment, required for voice features. Total stack cost: Under $20/month to run in production.

Challenges we ran into

Gemini quota limits during development. We hit rate limits constantly while building and testing. We solved this by switching from the Python SDK to direct REST API calls, building a retry loop across four model versions, and writing a rich keyword-based fallback so the app stays useful even when Gemini is unavailable. Deploying on Render's free tier. Render defaults to Python 3.14, which breaks pydantic-core because it tries to compile Rust source code on a read-only filesystem. Solution: pin PYTHON_VERSION=3.11 as an environment variable. Auth0 callback URL mismatches. Every time Vercel generated a new preview URL on a push, we had to update the Auth0 allowed callback URLs. We solved this permanently by setting a stable custom alias on Vercel. Merging two different codebases mid-hackathon. Two team members built separate versions of the product — the FinGuard financial health tool and the Life Impact Simulator. Merging them without breaking either meant creating separate models, separate routes (/api/analyze vs /api/simulate), separate onboarding flows, and separate dashboard pages — while sharing the same auth, TTS, and chat infrastructure.

Accomplishments that we're proud of

The shock test in the Life Impact Simulator. Nobody asked us to build it — we just thought, "the most important question isn't whether you can afford this decision today. It's whether you can survive this decision if something goes wrong." So we simulate a job loss across all three scenarios and show you exactly which path fails and why. That felt like real product thinking. We're also proud that the whole thing is actually deployed, live, and working. It's not a mockup. The backend is on Render. The frontend is on Vercel. Auth0 handles login. ElevenLabs reads responses aloud. A real person can go to the URL right now and get a real financial health assessment in under 3 minutes. And we're proud of the fallback chat system. When the Gemini quota runs out. which it often does, the AI Coach still gives smart, specific, useful answers across 9 financial topics. The product works regardless.

What we learned

We learned that the hardest part of building for financial accessibility isn't the technology. It's the language. Every early version of our copy used words like "liquidity," "liability exposure," and "coverage gap analysis." We had to rewrite everything at a Grade 8 reading level, multiple times. That discipline, speaking plainly, turned out to be the most important design decision we made. We also learned that deterministic models can be more trustworthy than AI for core financial logic. The Life Impact Simulator doesn't use AI to generate scores — it uses a carefully tuned mathematical model with real income brackets, employment stability factors, and scenario modifiers. AI is used for explanation and coaching. That separation made the product both more reliable and more debuggable. And practically: always pin your Python version on Render.

What's next for FinGuard AI

Real data integration. We'd add CFPB and FDIC public datasets to make flood risk, insurance cost estimates, and regional benchmarks more accurate. More life decisions in the simulator. Right now, we support 4 decisions: buy a car, move cities, take a loan, and change jobs. The next set would include having a child, going back to school, starting a business, and buying a home. SMS and WhatsApp channel. The AI Coach should work via text message. No app required, no login, no data plan beyond SMS. That's how you actually reach the gig worker or recent immigrant who needs this most. Multilingual support. ElevenLabs and Gemini both support multiple languages. A Spanish-language version is the highest-impact next step. Partnerships with community organizations. The product is most valuable when it's trusted. Partnering with credit unions, community colleges, and immigrant service organizations to deploy FinGuard as a free tool is how this reaches the people it was built for.