I already have a TradingView indicator written in Pine Script that fires fairly reliable entry and exit signals, yet its behaviour drifts whenever market regimes shift. I’d like you to plug machine-learning intelligence behind that script, back-test it thoroughly on both historical and real-time feeds, and harden the model so it stays robust through bull runs, crashes, and sideways stretches alike. You’ll be free to work in Python (TensorFlow, PyTorch, scikit-learn—whatever suits) as long as the finished logic ultimately drives orders generated from the existing Pine code. Real-time data access and broker execution pipelines are already in place; what I need is the learning layer plus a rigorous validation suite. Deliverables • Clean, reproducible back-test notebook covering at least 10 years of historical data and live forward tests • A trained model (supervised, unsupervised, or hybrid—your call) that demonstrably improves drawdown control and preserves or lifts overall expectancy • Refactored Pine Script or accompanying API calls that pipe the model’s output into TradingView in real time • Short hand-off report explaining hyper-parameter choices, retraining triggers, and how to extend the framework for new assets Acceptance criteria The bot must maintain its edge when volatility spikes, with no more than a 5 % drop in Sharpe ratio during live walk-forward tests compared with back-test results. If this aligns with your expertise in adaptive ML for financial markets, let’s talk specifics and set milestones.