MedCompress
Compression-as-clinical-context: condensing long medical documents into structured, faithful summaries that downstream LLM pipelines and decision-support workflows can actually consume.
Projects
Applied machine learning projects across computer vision, NLP, sequence models, and ML for health and climate, plus selected systems work.
Compression-as-clinical-context: condensing long medical documents into structured, faithful summaries that downstream LLM pipelines and decision-support workflows can actually consume.
End-to-end image captioning pipeline using VGG16 as a CNN encoder paired with an LSTM decoder, trained on Flickr 30k. CUDA-tuned for parallel batch processing, evaluated with BLEU, presented at the 47th Annual Beloit Student Symposium.
SOM-based unsupervised clustering of geographic regions by atmospheric pattern, layered with ANN and CNN classifiers to identify regions with the strongest climate change signal. Reported with confusion matrices on global maps.
Binary classification models using RNN and CNN on the UCI Heart Failure Clinical Records dataset, predicting mortality risk from clinical features. Compared with K-Means and SOM clustering to surface patient groupings before supervised training.
Text classification pipeline on the Kaggle Fake News dataset using tokenization, stopword removal, stemming, and TF-IDF vectorization. Compared Logistic Regression, Naive Bayes, and Random Forest classifiers across a held-out test set.
Full data mining workflow on an insurance cost dataset: exploratory analysis, feature engineering, and training of ANN, SVM, and Decision Tree regressors. Compared by RMSE and R² with feature-importance visualizations.
Python automation on Red Hat Enterprise Linux managing the full student account lifecycle in Active Directory, triggered by enrollment and departure events. REST API workflows synchronized account data with upstream HR and registrar systems; cron-scheduled Python pipelines generated lifecycle reports on a daily and weekly basis.
Supported Cisco switch firmware upgrades across state broadcast facilities, automated recurring reporting and billing workflows in Microsoft Power Automate, and produced technical network documentation including topology diagrams and runbooks.