SaaS Entrepreneurship and Technical Skills

Configure and verify a reproducible local Python environment with TensorFlow and version-controlled dependencies.

Transform market-data arrays using TensorFlow tensor shapes, dtypes, indexing, reshaping, and broadcasting.

Use tf.Variable and tf.GradientTape to compute gradients of a scalar loss.

Train a simple linear model with an explicit TensorFlow training loop.

Define a next-day target representing whether an Indian equity outperforms the NIFTY 50 benchmark.

Build a reproducible function that downloads, aligns, caches, and validates daily OHLCV data for selected Indian equities and the NIFTY 50.

Construct lagged return, volatility, volume, and market-regime features using only information available at prediction time.

Create chronological train, validation, and test partitions with preprocessing fitted exclusively on training data.

Build batched tf.data pipelines that preserve the chronological boundaries of the dataset partitions.

Build and compile a Keras multilayer perceptron for benchmark-relative binary classification.

Train a Keras model with checkpointing and early-stopping callbacks.

Interpret training and validation curves to distinguish underfitting from overfitting.

Establish a non-neural baseline against which to judge the TensorFlow model.

Apply L2 regularization and dropout to a Keras model to reduce overfitting.

Evaluate probability predictions using discrimination, calibration, and class-sensitive metrics.

Implement walk-forward validation without leaking future market observations into model development.

Construct rolling feature windows for a sequence model without crossing dataset boundaries.

Build and train a Keras recurrent network on multivariate equity sequences.

Compare the recurrent network with the multilayer perceptron under an identical evaluation protocol.

Save and reload the selected Keras model together with its fitted preprocessing artifacts.

Select a probability-to-position decision threshold using validation data only.

Backtest the selected long-or-cash signal after transaction costs on untouched test data.

Write automated checks for feature-target alignment, tensor shapes, and chronological split boundaries.

Publish a reproducible GitHub repository with a README and model card that document methodology, results, limitations, and non-advisory scope.