A Smarter Way to Reinvest Liquidity Pool Rewards: Using Machine Learning on Solana

I started by tracking Orca LP rewards on Solana and ended up building a contextual bandit that learns when to buy, sell, or hold SOL based on price patterns and past trade outcomes.

This project combines:

• A Flask web app
• Reinforcement learning (contextual bandits)
• Live price prediction
• SOL/USDC liquidity pool monitoring

It evolved from a passive analytics tool into a smarter system for simulating trades using real-time data and historical trends.

Liquidity pools like Orca’s SOL/USDC offer passive rewards, but I wanted to go one step further:
What if you reinvested those rewards into SOL using a machine-learning model that understood price context?

That idea led to building a trading simulator that:

• Learns from market indicators (Sharpe ratio, momentum, SMA)
• Evaluates the impact of each trade

• Tracks portfolio performance over time

• Python + Flask for the web app

• SQLite for local data persistence

• LiveCoinWatch API for SOL price data

• Helius API for Orca LP activity

• Contextual Bandits (via river or Vowpal Wabbit conceptually)

• Chart.js + Tailwind for the frontend

A contextual bandit is a type of reinforcement learning algorithm that chooses actions based on the current state (called “context”) and learns from the reward.

In this case:

• Context = features like price deviation from 24h low/high, rolling mean, Sharpe ratio, momentum, and current portfolio state
• Actions = buy, sell, or hold
• Reward = based on realized/unrealized profit, market timing, and trade quality

📈 Key Features

• Live portfolio tracking and trade logs

• Reward calculations that factor in profit, timing, and trend alignment

• Automatic SQLite logging for all trades and portfolio snapshots

• Model state saved between runs

• (Optional) Price predictor using rolling mean and recent volatility

🔍 Challenges & Tradeoffs

• The model needs time to learn. Early trades are often naive.
• There’s no “real” trading yet—only simulation.
• We’re still experimenting with the best reward functions.
• Future goal: add backtesting, deploy to simulate with live SOL rewards.

Try it!

GitHub: [https://github.com/JHenzi/OrcaRewardDashboard]

To run locally:

1. Clone the repo
2. Set up a .env file with your API keys
3. Run flask run
4. Visit http://localhost:5030 to explore the dashboard

🛣️ What’s Next

Next steps:

• Backtest with historical price data
• Refactor to make predictions available via API
• Possibly deploy as a hosted dashboard for live tracking
• Improve reward function based on more advanced trading signals

Let’s Talk?

Would love to hear feedback on:

• Reward strategy improvements
• Trading signal features to add
• How you’d use this if it supported real trading

Drop a comment below or reach out on GitHub!