PulseFi — Gen Z Banking Dashboard
Smart banking dashboard with FHI scoring, Gemini AI recommendations, and 2,000 synthetic Gen Z user profiles. Zero backend — runs entirely in the browser.
Most banking apps show your balance and call it a day. PulseFi goes further — it computes a Financial Health Index (FHI) score using a 6-step ML pipeline, assigns you a money personality, and surfaces AI-generated recommendations via Gemini Flash Lite. All of this runs entirely in the browser with no backend, no build step, and no npm.
The FHI engine is the brain: it takes 11 financial inputs, normalizes them into 8 feature scores (net worth ratio, DTI, savings rate, investment ratio, emergency fund coverage, spending ratio, spending volatility, panic sell tendency), applies Logistic Regression L1 coefficients to produce both a Baseline and Enhanced FHI score, then clusters the user into one of four money personality archetypes using K-Means with pre-computed centroids.
The dashboard has six fully wired screens — Home, Financial Health, Pulse AI, FHI Engine, Savings Goals, and Investments. Every financial action (send money, pay bills, top up, add to goals) updates balances in real time and silently recalculates the FHI score across all screens. It's designed for 18-30 year olds in Singapore, with investment behavior carrying 59% of the FHI weight based on research showing it's the strongest predictor of long-term financial resilience.
FHI Score 11.5 — Building Foundations. Cluster assignment: Developing. Building financial foundations. Some savings but limited investment coverage. Investment behavior accounts for 59% of the total FHI weight — the strongest predictor of long-term financial resilience for ages 18-30.
- ▸Zero-dependency frontend: vanilla JS, no React, no build tools. Charts rendered with Canvas API. Entire app is a single HTML file with modular JS files.
- ▸FHI pipeline: preprocessing → feature standardization → L1 logistic regression (baseline + enhanced) → XGBoost validation (3 models, AUC 0.67) → K-Means clustering (K=4, silhouette 0.225) → localStorage persistence.
- ▸AI recommendations pre-generated via batch Gemini Flash Lite calls against the full 2,000-user dataset, stored as static JSON — no API key needed at runtime.