Introducing AlphaEar Dashboard — AI-Powered Stock Forecasting

After weeks of building and iterating, I’m excited to share AlphaEar Dashboard — an open-source, AI-powered stock forecasting dashboard now live at alphaear.oilygold.xyz.

What Is AlphaEar Dashboard?

AlphaEar Dashboard is a self-hosted web application that provides rolling 5-day stock price forecasts for over 40 US and Australian stocks and ETFs. It uses linear regression on historical data to project near-term price movements, then tracks accuracy over time so you can see how well the predictions actually hold up.

Think of it as a quantitative research tool — not a crystal ball. It surfaces data, trends, and signals so you can make more informed decisions.

Core Features

🔮 5-Day Rolling Forecasts Every morning at 8 AM, the pipeline pulls fresh data from Yahoo Finance and generates new 5-day price projections. The dashboard shows predicted close prices alongside confidence bands.

📐 Predicted vs Actual Once the market closes, forecasted prices are compared against real closes. Green means the prediction landed within 2% — red flags larger deviations. This isn’t about being right every time; it’s about understanding when and why the model drifts.

📊 Accuracy Tracking MAPCE (Mean Absolute Percentage Close Error) is tracked per ticker over time. Interactive Chart.js graphs show whether accuracy is improving, degrading, or holding steady across different market regimes.

🔥 Market Movers A scanner runs every 4 hours to surface stocks with unusual volume and price movement — the hottest gainers and coldest losers at a glance.

📰 News & Sentiment Per-ticker news tabs pull Reddit discussions, Yahoo Finance headlines, and sector intelligence. The idea is to pair quantitative signals with qualitative context.

🌗 Dark/Light Mode + i18n The UI adapts to your system theme and supports both English and Traditional Chinese (繁體中文), with a glass-morphism design system built from scratch.

👥 Multi-User JWT-based authentication means everyone gets their own SQLite database, custom watchlists, and personalized forecast views.

Tech Stack

Layer	Technology
Backend	FastAPI (Python) — 30+ API endpoints
Frontend	Vanilla JS SPA with Chart.js
Database	SQLite (per-user)
Auth	JWT (register, login, token refresh)
Data Source	Yahoo Finance
Charts	Chart.js 4
Deployment	Railway

The entire stack is intentionally lightweight — no React, no ORM, no Docker. A single FastAPI process serves both the API and the static dashboard, keeping the architecture simple and the resource footprint tiny.

Architecture at a Glance

Yahoo Finance API
       │
       ▼
generate_forecasts.py  ──→  /tmp/kronos_forecasts.json
       │
       ▼
  app.py (FastAPI)     ──→  serves /api/live + dashboard
       │
       ▼
scripts/ingest.py      ──→  per-user SQLite databases

Three cron jobs keep the system fresh:

Forecast generation runs daily at 8 AM
Accuracy recalculation runs daily at 6:30 AM
Market movers refresh every 4 hours

Tracked Tickers

Right now the system covers a curated set across markets:

US Tech: AAPL, NVDA, TSLA, MSFT, GOOGL, AMZN, META, AMD, and more
US ETFs: SPY, QQQ, IWM, DIA
AU Stocks: BHP, CBA, CSL, RIO, NAB, ANZ, WBC, WOW, and more
AU ETFs: IVV, VAS, VGS, VDHG, NDQ, DHHF, IOO
Global: TSM, ASML

Custom tickers can be added from the Settings panel — the system fetches live data and generates forecasts on the fly.

Ensemble Models

Beyond the core linear regression pipeline, AlphaEar also supports an ensemble mode that combines multiple model approaches:

Kronos — the default linear regression model
Prophet — Meta’s time-series forecasting library
XGBoost — gradient-boosted trees

The ensemble predictor runs these in parallel and weights their outputs, aiming to smooth out individual model weaknesses. It’s available via make run-ensemble for anyone who wants to experiment.

Why I Built This

I wanted a tool that:

Surfaces quantitative data without the noise — no hype, no “buy now” alerts, just projections and accuracy metrics
Tracks its own mistakes — the accuracy tab exists because no model is perfect, and seeing where predictions fail is more valuable than pretending they won’t
Runs on minimal infrastructure — the entire thing fits on a single Railway instance with a SQLite database
Is open source — anyone can fork it, run it locally, or deploy their own instance

Try It Yourself

The dashboard is live at alphaear.oilygold.xyz — register an account and explore the forecasts.

The code is open source on GitHub: github.com/acchuang/alphaear-dashboard

git clone https://github.com/acchuang/alphaear-dashboard.git
cd alphaear-dashboard
make install
make run

Visit http://localhost:5050 and you’re up and running.

What’s Next

A few things on the roadmap:

Backtest validation — historical replay to measure how forecasts would have performed in past market conditions
More data sources — beyond Yahoo Finance, pulling from additional providers for redundancy
Alert system — configurable notifications when forecasts cross user-defined thresholds
Mobile PWA — progressive web app support for a proper mobile experience

⚠️ AlphaEar provides quantitative and technical analysis only. This is not financial advice. Stock markets are inherently unpredictable — past accuracy does not guarantee future performance.