Wise Wallet

Why track your expenses by hand? Wise Wallet brings AI into your finances to predict, analyze, and help you make smarter money moves.

Comment

Wise Wallet 🧠💸

Your AI-powered voice + web assistant for smart financial decisions.

🚀 Inspiration

Money stress is real — especially when it sneaks up on you. We wanted to build something that not only tracks your expenses, but actually helps you understand and control them before the damage is done. Our team talked about how most tools just dump raw numbers at you. We wanted something smarter. More helpful. Something that talks back and gives you actual insights — almost like a financial coach, right inside your browser and even your kitchen (thanks, Alexa).

🤖 What it does

Wise Wallet is a two-part system:

  • Web App – A beautiful dashboard that predicts your end-of-month expenses, classifies your spending into needs and wants, and visualizes it all so you can take action early.
  • Alexa Integration – A voice-based assistant that answers financial questions in real-time using your latest expense data, processed by an LLM (Google Gemini) and backed by our own classification and regression ML models.

Key Features:

  1. Smart Voice Insights via Alexa Ask, “What did I spend on food last week?” or “Am I spending too much on wants this month?” — Alexa responds using real-time data and custom-tuned prompts optimized for relevant, helpful answers.

  2. Expense Prediction (Web App) We predict your total monthly expenses using a trained Random Forest Regressor. So if it's only the 16th, you still get a smart guess of how the month is shaping up, visually shown in a sleek graph.

  3. Spending Breakdown & Categorization (Web App) All expenses are analyzed using a custom-trained Random Forest Classifier to determine whether each entry is a “need” or a “want.” This breakdown is shown in a semi-circular graph (think doughnut chart, but cooler), alongside a scrollable list of all expenses.

🛠️ How we built it

  • Frontend: React + TypeScript + TailwindCSS for fast, modern UI development.
  • Backend: AWS Lambda microservices written in Python, serving ML models trained with scikit-learn.
  • Voice Interface: Alexa Skills built using Node.js, with dedicated Lambda functions that interface with our APIs and trigger Google Gemini prompts for deep LLM insights.

  • Machine Learning:

    • Classifier: Trained on curated financial transaction data to separate “needs” vs “wants.”
    • Regressor: Predicts monthly totals using Random Forest Regression trained on historic personal finance patterns.

  • Infra: AWS (API Gateway + Lambda + S3), designed for scalability and cost-efficiency.

😬 Challenges we ran into

  • LLM prompt tuning is art, not science. Getting Gemini to respond with the right context without hallucinating or being too vague took a LOT of tweaking.
  • Data was messy. We had to simulate and clean large sets of expense data, since real user data wasn't available for training.
  • Latency with Alexa. Real-time voice interactions with backend inference models needed optimization. At first, we were facing ~5s response times. We fixed it by caching common intents and reducing cold starts.

🏆 Accomplishments that we're proud of

  • Pulled off multi-modal AI (voice + web + ML + LLMs) in one coherent product.
  • Built and deployed a working system with real ML inference, not just hardcoded logic.
  • Created a natural and actually useful voice assistant experience — not just gimmicky.
  • Learned a TON about end-to-end ML deployment and Alexa skill development.

📚 What we learned

  • Prompt engineering is low-key one of the most powerful — and overlooked — tools when integrating LLMs.
  • ML is nothing without usable data. We had to learn how to simulate, clean, and validate data fast.
  • Voice UIs are a whole different beast. Timing, tone, and error handling matter way more than on a web app.
  • AWS Lambdas rock — when used properly. Cold starts and timeouts are real, but solvable.

🔮 What's next for Wise Wallet

  • Bank Integration: Hook up with Plaid or open banking APIs to sync real bank data instead of mock sets.
  • Mobile App: Extend the web dashboard into a cross-platform React Native app.
  • More LLM Use: Add a daily “Financial Digest” feature — a quick morning voice/audio summary of yesterday’s spending and today’s predicted trends.
  • Finetuned LLM model: Eventually train a domain-specific LLM on financial Q&A behavior to improve accuracy and reduce prompt overhead.
Share this project:

Updates