DEV Community

ik park
ik park

Posted on • Originally published at nextcandle.io

I Built an AI That Finds 99% Similar Historical Candlestick Patterns — Here's What I Learned

#ai #machinelearning #crypto #trading

The Problem

Every trader has stared at a chart and thought: "I've seen this pattern before." But human memory is unreliable, and manually searching through years of historical data is impractical.

So I built NextCandle — a deep learning tool that does exactly this: takes the current candlestick pattern and finds the most similar historical patterns, then shows what happened next.

How It Works

  1. Pattern Extraction: Captures the current N candles (configurable from 3 to 100) from any supported asset
  2. Deep Learning Similarity: Uses a trained model to compare against all historical patterns
  3. Top-K Matching: Returns the top 3 most similar patterns with similarity scores (often 95%+)
  4. Outcome Analysis: Shows the MAX/MIN price movement that followed each historical pattern
  5. Trend Score: Aggregates results into a 0-100 bullish/bearish score

What Makes It Different

  • Not prediction — it's historical pattern retrieval. "This happened before, here's what followed."
  • Objective data — no opinions, no bias, just statistical facts
  • Multi-asset: BTC/USDT, ETH/USDT, NASDAQ, S&P 500, Gold
  • Multi-timeframe: 1H, 4H, 1D, 1W
  • Extended lookback: Up to 100 candles for macro pattern recognition

Example Output 

BTC/USDT (4H) | 3 Candles | 2026.02.22

Top 1: 2025.03.22 — 99.1% similarity
  → MAX: +2.3%  MIN: -0.5%

Top 2: 2024.11.15 — 97.8% similarity  
  → MAX: +1.8%  MIN: -1.2%

Top 3: 2023.08.03 — 96.5% similarity
  → MAX: +3.1%  MIN: -0.3%

Trend Score: 85/100 (Bullish)
Enter fullscreen mode Exit fullscreen mode

The Tech Stack

  • Deep learning model trained on years of OHLCV data
  • Real-time data feeds for supported assets
  • Automated analysis every 4 hours
  • Free Telegram channel with instant updates

Try It

🌐 Web: nextcandle.io
📱 Telegram (free 4H updates): t.me/nextcandle_official

Currently supporting:

  • BTC/USDT, ETH/USDT
  • NASDAQ, S&P 500, Gold
  • Multiple timeframes (1H, 4H, 1D, 1W)

What's Next

  • More assets and exchanges
  • Custom alerts for specific pattern matches
  • API access for algorithmic traders
  • Backtesting integration

Disclaimer: This is a data analysis tool, not financial advice. Past patterns don't guarantee future results. Always do your own research.

Would love to hear your thoughts — especially from anyone doing quantitative or fractal-based trading. How do you use historical pattern data in your strategies?

Top comments (0)