ChainPulse

Cryptocurrency sentiment analysis dashboard powered by AI

ChainPulse aggregates crypto news, analyzes sentiment using FinBERT, and delivers actionable market insights through interactive visualizations. This platform is made for people that want to read the market fast.

Click the button below for the LIVE DEMO

---

✨ Features

• AI-Powered Sentiment Analysis - FinBERT analyzes 100+ daily crypto articles with 85%+ accuracy
• Market Sentiment Tracking - Aggregated scores across 24h, 7d, 30d time periods
• Interactive Visualizations - D3.js bubble charts showing sentiment across 60+ coins
• Smart Coin Matching - Automatic extraction and tagging of mentioned cryptocurrencies
• Performance Optimized - Redis caching + batch processing for sub-400ms responses
• Duplicate Prevention - Intelligent deduplication ensures clean news feed

---

Tech Stack & Tools Used

---

System Design

Architecture Diagram

Key Design Decisions and Assumptions

1. Batch Processing for ML Inference

• Processes 16 articles simultaneously instead of sequential analysis
• Reduces sentiment analysis time from 3 minutes to 15-20 seconds (~10x faster)

2. Caching Strategy

• 5-minute Redis TTL balances freshness with performance
• Query-level caching for news, sentiment, and coin aggregations
• TanStack Query for optimistic frontend updates

3. Database Optimization

• Indexed published_at and sentiment_label for fast filtering
• Eager loading (joinedload) eliminates N+1 query problem
• Unique constraint on (title, published_at) prevents duplicates

4. Startup Model Preloading

• BERT model loads during app startup (not first request)
• Singleton pattern prevents redundant model instances

Performance Optimizations

Optimization	Before	After	Impact
Batch BERT Processing	180s	20s	9x faster
Database Indexes	2000ms	300ms	6.7x faster
N+1 Query Fix	1500ms	150ms	10x faster
Redis Caching	DB query	<50ms	30x faster

---

Screenshots

Dashboard Overview

_Sentiment gauge, market trends

Leaderboard and Interactive Bubble Chart

D3.js visualization showing sentiment distribution across 60+ cryptocurrencies

News Feed with Sentiment & Their Related Coins

AI-analyzed crypto news with sentiment scores and coin tags

---

Quick Start

Prerequisites

• Node.js 18+
• Python 3.11+
• PostgreSQL 14+
• Redis (or Upstash account)

Backend Setup

cd backend
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt

# Configure environment
cp .env.example .env
# Edit .env with your credentials

# Run migrations
alembic upgrade head

# Start server
uvicorn app.main:app --reload

Frontend Setup

cd frontend
npm install

# Configure environment
cp .env.local.example .env.local
# Edit .env.local with API URL

# Start development server
npm run dev

Visit http://localhost:3000

---

📚 API Documentation

Endpoints

News

• GET /api/news - Paginated news with filters (period, sentiment, search)
• POST /api/news/refresh - Fetch latest news from CryptoPanic

Sentiment

• GET /api/sentiment/aggregate - Overall market sentiment by period
• GET /api/coins/sentiment - Top 5 bullish/bearish coins
• GET /api/coins/bubble - All coins data for bubble chart

---

🥽 Technology Deep Dive

Natural Language Processing (NLP)

• FinBERT: Financial domain-specific BERT model fine-tuned on financial news
• Batch Inference: Processes multiple texts simultaneously for 10x speedup
• Normalization: Scores mapped to -1 (bearish) to +1 (bullish) scale

Coin Matching Algorithm

• Aho-Corasick: Multi-pattern matching for efficient coin symbol detection
• Levenshtein Distance: Fuzzy matching for misspellings and variations
• Coverage: 70+ major and most popular cryptocurrencies (BTC, ETH, SOL, etc.)

Data Pipeline

1. Ingestion: CryptoPanic API provides 20-30 articles per fetch
2. Analysis: FinBERT batch processing (16 articles/batch)
3. Extraction: Coin matcher identifies mentioned cryptocurrencies
4. Storage: PostgreSQL with deduplication and indexing
5. Caching: Redis stores aggregated results (5-min TTL)

---

Acknowledgments & References

• ProsusAI/finbert - Financial sentiment analysis model
• CryptoPanic - Crypto news aggregation API
• Railway - Backend deployment
• Vercel - Frontend deployment

---

Made by Andrew Tedjapratama