Finn — Personal Finance & Market Forecasting
A full-stack personal-finance and market-analytics app: track instruments, build watchlists and portfolios, run statistical and ML forecasts with walk-forward validation, review risk, and read FinBERT news sentiment — with LATAM coverage (IPSA + Chile macro) behind a fast, server-rendered UI.
Business Context
Retail and analyst-facing finance tools tend to split into two camps: polished dashboards with shallow analytics, or powerful quant libraries with no usable interface. Neither serves someone who wants real forecasting and risk analysis, honest backtesting, and coverage of Latin American markets — which mainstream tools largely ignore — without standing up a research environment from scratch.
Strategic Value
Finn packages a research-grade analytics stack behind a fast, server-rendered UI. Its differentiators are rigor and reach: forecasts are scored with walk-forward cross-validation (train on the past, test on the unseen next window) instead of leaking the future into the fit, and backtests charge transaction costs so a strategy has to survive contact with reality. FinBERT brings finance-aware sentiment — distinguishing "missed guidance" from "raised guidance," which a general sentiment model fumbles — and first-class LATAM coverage (IPSA, Chilean macro, BCCh) makes it usable for the markets it was built in. Risk analytics (volatility, Sharpe, max drawdown, VaR, efficient-frontier optimization, regime detection) and advanced Plotly visualizations (fan charts, drawdown, fundamentals bubbles, regime shading) round it out, with per-user persistence behind auth.
The Challenge
Most quant demos default to US tickers and frictionless backtests — they look impressive and mislead. A credible market-analytics tool has to validate forecasts the way they would actually be traded, charge realistic costs, score news with finance-aware models, and cover the markets the user actually lives in.
Our Approach
A server-rendered FastAPI + HTMX app over a quant stack (pandas, numpy, scipy, statsmodels, scikit-learn). Forecasting spans ARIMA/SARIMA, exponential smoothing, Prophet-style decomposition, and gradient-boosted regressors, evaluated with walk-forward cross-validation. Risk covers volatility, Sharpe, max drawdown, VaR, mean-variance optimization, and regime detection. FinBERT (HF Inference API) scores headline sentiment; LATAM coverage adds the IPSA universe, Chilean macro series, and a Banco Central de Chile stub. Auth-backed watchlists and portfolios persist per user.
Key Performance Indicators
| KPI | Baseline | Result | Impact |
|---|---|---|---|
| Forecast Validation | In-sample fit (future leakage) | Walk-forward cross-validation | Out-of-sample error you can trust |
| Market Coverage | US tickers only | IPSA + Chile macro + BCCh | Usable for LATAM markets |
Architecture
finn architecture
A Quant Demo That Doesn’t Lie to You
Finn is a full-stack personal-finance and market-analytics web app: track instruments, build watchlists and portfolios, run forecasts, and review risk — all behind a clean, fast, server-rendered UI. What separates it from the average finance dashboard is what it refuses to fake.
Forecasting You Can Trust
Finn runs several forecasting models — ARIMA/SARIMA, exponential smoothing, Prophet-style decomposition, and gradient-boosted regressors — but the important part is how they’re evaluated. Walk-forward cross-validation trains on the past and tests on the unseen next window, rolling forward, so the reported error is genuinely out-of-sample. No future leaks into the fit. Confidence intervals and scenario fan charts make the uncertainty explicit.
Risk, Portfolios, and Honest Backtests
Build portfolios, track allocations, and compute returns with a full risk battery: volatility, Sharpe, max drawdown, VaR, mean-variance optimization (efficient frontier), and regime detection. Backtests charge transaction costs — the fastest way to kill a strategy that only looked good because it traded for free.
Finance-Aware Sentiment + LATAM Coverage
Financial sentiment is its own dialect: “missed guidance” and “raised guidance” sit next to each other in vocabulary space but mean opposite things to a price. Finn scores headlines with FinBERT (HF Inference API) instead of a general sentiment model. And rather than defaulting to US tickers, it adds first-class LATAM coverage — the IPSA universe, Chilean macro series, and a Banco Central de Chile stub — so it’s a tool for the markets it was built in.
Architecture
A FastAPI + HTMX server-rendered application over a quant stack (pandas, numpy, scipy, statsmodels, scikit-learn), with SQLAlchemy/Alembic persistence and interactive Plotly charts (fan charts, drawdown, fundamentals bubbles, regime shading). Auth-backed watchlists and portfolios persist per user. Live at finn.fasl-work.com.
Technology Stack
Visual assets for this project are not publicly available.