Most "RSI bots" buy every dip and die in bear markets. This one doesn't. It stacks three timeframes of confirmation — daily macro trend, hourly order flow, 5-minute RSI execution — so it only fires when every layer agrees. Risk is bounded by a 2× ATR dynamic stop, and the whole thing is backtested through a Streamlit dashboard that spits out Sharpe, CAGR, and drawdown like a hedge fund tear sheet.

Four layers of filtering. A trade only happens when all four line up:

The whole idea: stop trying to catch falling knives in a crypto winter. Wait for the macro tape to confirm, then scalp the micro dips that real institutions are buying.

• Multi-timeframe by design — not "RSI on a 5-min chart and pray." The strategy is literally impossible to enter during a sustained downtrend.
• Order-flow aware — uses Binance taker buy/sell volume as a leading indicator, not just price.
• Institutional metrics — CAGR, Sharpe, Max Drawdown, Expectancy, Win Rate. Not just "did number go up."
• Decoupled architecture — src/engine.py is the strategy core, app.py is the dashboard. Swap either independently.
• Config-driven — every threshold lives in config.yaml. Tune without touching code.

# 1. Clone
git clone https://github.com/Kiankk/BTC-RSI-Bot.git
cd BTC-RSI-Bot

# 2. Install (minimal deps — pandas, pandas-ta, streamlit, plotly, pyyaml)
pip install -r requirements.txt

# 3. Drop your OHLCV data
#    Expected: data/btc_1m_orderflow.csv
#    Columns:  Open time, Open, High, Low, Close, Volume, Taker buy base asset volume
#    (Direct download from Binance Vision works out of the box.)

# 4. Run the dashboard
streamlit run app.py
# → opens at http://localhost:8501

Everything lives in config.yaml:

strategy:
  initial_capital:     500.0    # Starting equity ($)
  position_size:       500.0    # Fixed per-trade size
  fee_rate:            0.0006   # 0.06% per side (Binance taker)
  hard_stop_loss_pct:  0.015    # 1.5% hard stop

indicators:
  rsi_length:      14
  rsi_oversold:    40   # entry trigger
  rsi_overbought:  70   # exit trigger
  ema_length:      50

• 🏆 Strategy Leaderboard — ranks every variant by Sharpe, Net Profit, and Win Rate
• 📈 Equity Curve — interactive Plotly chart of cumulative balance
• 📉 Underwater Plot — drawdown over time, the chart that actually tells you if a strategy is survivable
• 📊 Trade Distribution — P&L histogram
• 📋 Trade Log — exportable CSV of every entry/exit with timestamps

BTC-RSI-Bot/
├── data/                   # Your OHLCV data (gitignored)
├── src/
│   ├── engine.py           # DataLoader · FeatureFactory · StrategyEngine
│   └── logger.py           # Production logging
├── tests/
│   └── test_engine.py      # Pytest suite
├── app.py                  # Streamlit dashboard
├── config.yaml             # All tunable parameters
└── requirements.txt

2026-03-25 15:32:10 │ QuantEngine │ Loading Raw Data from data/btc_1m_orderflow.csv...
2026-03-25 15:32:11 │ QuantEngine │ Loaded 142,560 rows.
2026-03-25 15:32:12 │ QuantEngine │ Engineering Multi-Timeframe Matrix...
2026-03-25 15:32:15 │ QuantEngine │ Running V21 Strategies (Macro Filter + Fixed Sizing)...
2026-03-25 15:32:18 │ QuantEngine │ ✓ Hybrid_V21          Sharpe 1.84  CAGR 47.2%  MDD -11.3%
2026-03-25 15:32:18 │ QuantEngine │ ✓ Hybrid_V21_NoFlow   Sharpe 1.21  CAGR 28.6%  MDD -19.8%
2026-03-25 15:32:18 │ QuantEngine │ ✓ Hybrid_V21_NoMacro  Sharpe 0.43  CAGR  8.1%  MDD -41.2%

The macro filter alone cuts max drawdown from -41% to -11%. That's the whole thesis.

• Live Binance execution via ccxt (paper mode first)
• Walk-forward optimization (no more in-sample curve fitting)
• Monte Carlo robustness tests
• Slack / Telegram alerts on signal generation
• Docker image for one-command deploy
• Extend to ETH, SOL, and the rest of the majors

This is a research project. Past backtest performance does not predict future results. Crypto can and will erase your account if you size positions like a degenerate. Run it in paper mode. Do your own research. Not financial advice.