SentimentEdge

Market sentiment dashboard that pulls financial news, runs it through a fine-tuned DistilBERT model, and shows how sentiment tracks against price movement. Also has a backtester so you can see how trading on sentiment signals would've played out.

Works in demo mode out of the box (no API keys needed).

Setup

You'll need Python 3.9+ and Node 18+.

Backend:

cd backend
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
cp ../.env.example .env   # optionally add a Finnhub API key
python app.py

First run downloads the FinBERT model (~400MB) and seeds 30 days of demo data.

Frontend:

cd frontend
npm install
npm run dev

Dashboard runs at localhost:5173, backend at localhost:5000. Vite proxies API calls automatically.

Training your own model (optional):

cd backend
python train/train_model.py

Fine-tunes DistilBERT on Financial PhraseBank (~5 min on CPU). Saves to backend/trained_model/ and gets picked up on next restart.

What's in here

• Sentiment vs price chart (dual axis, shows correlation over time)
• Live news feed with color-coded sentiment scores
• Accuracy tracker — did yesterday's sentiment predict today's price direction?
• Backtester with 3 strategies: threshold, momentum, and contrarian
• Paste-a-headline analyzer for quick sentiment checks

API

All under /api:

• GET /health — status + model info
• GET /dashboard?symbol=AAPL&days=30 — main dashboard payload
• GET /headlines, /sentiment/trend, /prices, /accuracy — individual data
• POST /backtest — run a strategy simulation
• POST /analyze — classify a single headline
• POST /fetch-now — trigger data refresh

Stack

• ML: PyTorch + HuggingFace Transformers (DistilBERT)
• Backend: Flask, SQLAlchemy, APScheduler, SQLite
• Frontend: React, TypeScript, Vite, Tailwind v4, Recharts
• Data: Finnhub API for news + prices