Looks like you have JavaScript disabled. For the full Twine experience, you will need to re-enable it.

Data science graduate student building machine learning models to answer practical business questions—like predicting loan defaults, classifying images, and generating text—and explaining results in plain language for business decision-makers. Experienced across the full lifecycle: problem framing, data preparation, model building and validation, and clear data storytelling with explainability tools such as SHAP.…Data science graduate student building machine learning models to answer practical business questions—like predicting loan defaults, classifying images, and generating text—and explaining results in plain language for business decision-makers. Experienced across the full lifecycle: problem framing, data preparation, model building and validation, and clear data storytelling with explainability tools such as SHAP.

Shreyash Sanjay Agrawal

Data Scientist, Data Engineer, AI Engineer, +2





Data science graduate student building machine learning models to answer practical business questions—like predicting loan defaults, classifying images, and generating text—and explaining results in plain language for business decision-makers. Experienced across the full lifecycle: problem framing, data preparation, model building and validation, and clear data storytelling with explainability tools such as SHAP.…Data science graduate student building machine learning models to answer practical business questions—like predicting loan defaults, classifying images, and generating text—and explaining results in plain language for business decision-makers. Experienced across the full lifecycle: problem framing, data preparation, model building and validation, and clear data storytelling with explainability tools such as SHAP.

Available to hire

Data science graduate student building machine learning models to answer practical business questions—like predicting loan defaults, classifying images, and generating text—and explaining results in plain language for business decision-makers.
Experienced across the full lifecycle: problem framing, data preparation, model building and validation, and clear data storytelling with explainability tools such as SHAP.

Skills

Experience Level

Expert

Expert

Expert

Expert

Expert

Expert

Work Experience

Text Generation with Neural Networks (Project)

September 1, 2025 - November 30, 2025

Built a character-level neural language model using the full text of Pride and Prejudice. Compared multiple architectures (single/2-layer LSTM, GRU, and regularized LSTM) to measure trade-offs between accuracy, training cost, and overfitting; best model achieved 2.99 validation perplexity. Evaluated output quality using complementary metrics including perplexity, BLEU, ROUGE, and diversity scores, and implemented a temperature-controlled sampler to balance creativity and coherence.

Image Classification with Deep Learning (Project)

September 1, 2025 - November 30, 2025

Trained a convolutional neural network to recognize objects across 60,000 images. Diagnosed failure modes (e.g., cat vs. dog confusion) and improved generalization by addressing overfitting using a deeper architecture, dropout, weight regularization, and image augmentation. Used Keras Tuner to perform systematic hyperparameter search, improving test accuracy from 70.4% to 80.1% (~10 point gain).

Credit Risk Prediction (Project)

January 1, 2025 - April 30, 2025

Built an XGBoost/Scikit-learn model to predict loan customers likely to default using 1M+ records and 54 engineered behavioral signals. Achieved 94% accuracy on customers the model had never seen. Added explainability with SHAP so flagged accounts can be justified to business users, customers, and regulators. Validated reliability using leakage-proof customer-level splits and two independent holdout sets to reflect production behavior. Simulated lending strategies using model scores and projected up to a 73% revenue lift while keeping default rates under 7%. Packaged the final model for deployment.