SentimentFX

Full-stack crypto sentiment intelligence platform. Scrapes financial news headlines, scores them with FinBERT (a finance-specific transformer model), and correlates sentiment against live GBP crypto prices to surface predictive signals.

Live: sentimentfx.org · app.sentimentfx.org · developers.sentimentfx.org

---

What it does

• Scrapes headlines every hour from GNews API + CoinTelegraph/CoinDesk RSS feeds
• Scores each headline with FinBERT (positive_prob - negative_prob → range -1 to +1)
• Fetches daily GBP prices via yfinance
• Computes Pearson correlation across 0–7 day lags to find the best predictive signal
• Exposes a public metered API with Stripe billing for developers

Tech Stack

Layer	Technology
Backend	Python 3.11, FastAPI, SQLAlchemy
AI Model	FinBERT (ProsusAI/finbert)
Database	PostgreSQL
Auth	Supabase (email + Google OAuth)
Payments	Stripe (subscriptions + metered API billing)
Frontend	React, Recharts, Vite
Data	GNews API, feedparser (RSS), yfinance
Email	Resend
Deploy	Railway (backend), Vercel (frontend)

Features

Dashboard

• Sentiment vs price chart with 7/30/90/ALL day ranges
• Pearson correlation with lag analysis (0–7 days)
• Momentum vs contrarian signal detection
• Paginated headline feed with per-article sentiment scores
• GBP/USD currency toggle
• CSV export (Pro)

Tickers: BTC, ETH, SOL, XRP, DOGE

Access tiers

• Free — BTC only, 30 day history
• Pro (£11.99/mo or £99.99/yr) — all tickers, full history, CSV export, sentiment alerts

Developer API (api.sentimentfx.org/v1/)

• /v1/sentiment/{ticker} — scored headlines
• /v1/summary/{ticker} — daily averaged sentiment
• /v1/prices/{ticker} — daily GBP closing prices
• /v1/correlation/{ticker} — Pearson r across lag periods
• Metered billing at £0.01/call after 100 free calls
• Rate limited via slowapi

Architecture

GNews API + RSS feeds
        ↓
   scraper.py → FinBERT (sentiment.py) → headlines table
                                              ↓
yfinance → prices.py → prices table    /dashboard/{ticker}
                                              ↓
                                       React + Recharts

APScheduler runs scrape_all() on a cron trigger at the top of every hour.

Getting Started

Backend

cd backend
python -m venv venv
source venv/Scripts/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt
cp .env.example .env          # fill in your keys
python -m uvicorn app.main:app --reload

Frontend

cd frontend
npm install
npm run dev

Sentiment Scoring

Uses FinBERT's output probabilities directly:

score = positive_probability - negative_probability

Range: -1 (maximally negative) to +1 (maximally positive). Neutral headlines score near 0. This avoids the label × confidence collapse that zeros out neutral predictions.

API Reference

Full docs at developers.sentimentfx.org

# Generate a free API key
curl -X POST https://api.sentimentfx.org/api/keys/generate \
  -H "Content-Type: application/json" \
  -d '{"email": "your@email.com"}'

# Use the key
curl https://api.sentimentfx.org/v1/summary/BTC \
  -H "X-API-Key: sfx_your_key_here"

Deployment

• Backend: Railway via Procfile + nixpacks.toml. Starts with uvicorn app.main:app --host 0.0.0.0 --port $PORT.
• Frontend/Landing/Docs: Vercel, auto-deploys from main.