Ten years of 1-minute stock data and a custom Pine Script indicator from TradingView are ready for you to turn into a robust signal-generating engine. The dataset must first be converted to Renko bricks; from there, I want both supervised and unsupervised deep-learning approaches explored so we can isolate the technique that flags high-probability entry and exit points most reliably. The core need is model generation. You’ll select or design the architecture—LSTM, Transformer, autoencoder hybrids, or any framework that proves effective—then train, validate, and benchmark it. I can supply continuous data feeds for additional walk-forward testing once an initial model is in place. You’ll run the experiments on your own GPU or on AWS you provision; if a different cloud or on-prem solution will accelerate training, outline that and we’ll switch. A TradingView account is essential so you can cross-check output on live Renko charts and refine the indicator rules. Deliverables • Cleaned & feature-engineered Renko dataset • Training notebooks / scripts with reproducible environments (Python, TensorFlow/PyTorch, etc.) • Trained model files plus lightweight inference script or API endpoint • Brief technical report summarizing methodology, metrics, and how to interpret the generated signals within TradingView Acceptance criteria • Model predicts buy/sell signals on out-of-sample data with clearly documented metrics (precision, recall, or custom profit factor) • All code executes end-to-end from raw CSV + Pine indicator to final predictions on a fresh machine using provided instructions • Visual overlay of predicted signals on a TradingView Renko chart matches the report’s examples Ready to move fast as soon as we agree on the initial architecture and toolset.