I'm a second year BTech student in Artificial Intelligence and Data Science. I like taking a messy dataset and turning it into something that actually works, from the first look at the data to a model I can explain and a demo someone else can click. Right now I'm moving from sequence models toward Transformers.
The card above rebuilds itself automatically from the GitHub API, so it is always current.
| Project | Problem | Approach | Highlight |
|---|---|---|---|
| ecg_data | Detecting abnormal heartbeats in ECG signals | LSTM, Bi-LSTM and XGBoost compared side by side, built with TensorFlow and Keras | Live Streamlit demo |
| predictive_maintenance | Estimating the Remaining Useful Life of aircraft engines from NASA C-MAPSS data | XGBoost and LSTM models | Sequence modeling on engine degradation data |
| mnist_dataset_model | Digit classification | Fully connected network tuned with a Keras Tuner Hyperband search | About 98.58% test accuracy, reproducible training pipeline, confusion matrix analysis |
| exp_learning_telco_churn | Predicting telecom customer churn | Logistic Regression and Random Forest, plus an ANN pipeline | Classical ML and deep learning compared on one problem |
| PremiumPulse | Predicting medical insurance charges | Ten regression models, from linear regression to XGBoost | Systematic hyperparameter tuning and reusable scikit-learn pipelines |
| beyond_accuracy | Predicting stroke risk from imbalanced patient data | Logistic regression with SMOTE oversampling | Classification threshold optimization |
- Compare, don't guess. I try more than one model on the same problem, so the choice is backed by results and not by habit.
- Look past accuracy. Class imbalance, thresholds and error types matter more than one headline number.
- Build pipelines, not one-off notebooks. Reusable preprocessing and training steps make the work easy to rerun and easy to check.
ANN, then CNN, RNN, LSTM, GRU, Encoder-Decoder and Attention. Now working through Transformers.
| Layer | Tools |
|---|---|
| Data | pandas, NumPy, seaborn, matplotlib |
| Classical ML | scikit-learn, XGBoost |
| Deep learning | TensorFlow, Keras, Keras Tuner |
| Demos | Streamlit |
| Environment | Python, Jupyter Notebook |
class Yugamdeep:
def __init__(self):
self.role = "AI and Data Science student, second year"
self.focus = ["applied ML", "deep learning", "data analysis"]
self.currently_learning = "Transformers"
def what_i_do(self):
return "Turn a messy dataset into a working, explainable pipeline"
print(Yugamdeep().what_i_do())