A selection of ML, NLP, and systems projects built during graduate study and independent research.
Transformers, Pyteomics
Found 73.6% and 95.8% prediction instability in Casanovo and π-PrimeNovo under physics-preserving spectral perturbations, revealing fundamental reliability gaps in de novo sequencing models and raising questions about their deployment in clinical proteomics.
CodebaseTransformers, PEFT, BitsAndBytes, LoRA
Processed unstructured text using private LLMs hosted locally for data privacy. Applied PEFT (LoRA) for efficient fine-tuning of summarizers and classifiers, boosting performance through prompt engineering without full-model retraining.
CodebasePyTorch, Scikit-learn, XGBoost, tsfresh
Built a predictive XGBoost model to forecast intracerebral hemorrhage (ICH) outcomes using time-series ICU data and PubMed radiology reports. Achieved 74.11% accuracy and 75.51% precision, aiding hospital decisions by identifying high-risk patients.
CodebaseHuggingFace Transformers, NLTK
Investigated transformer models for text summarization, comparing BERT, GPT-2, T5, and BART on medical report summarization. Addressed input size limits and domain-specific vocabulary challenges to improve summarization fidelity.
CodebasePyTorch, VGG-19
Implemented neural style transfer using a pretrained VGG-19 convolutional network to extract and disentangle content and style representations, generating a target image that combines the content of one image with the artistic style of another.
Codebase