Developing a temporal multimodal model (CT scans, EHR data, & radiology reports) with cross-attention and dynamic gating for forecasting disease progression & predicting hospital readmission using the INSPECT dataset.

• Building a scalable LMS platform with a RAG-powered AI assistant for 400+ users using AWS services (DynamoDB, S3, Cognito, API Gateway, Bedrock) and Pinecone with secure authentication and real-time progress tracking.

• Cleaning and organizing Excel datasets, creating workflow automations in Power Automate to improve data & team accuracy and efficiency.

• Evaluating 5+ LLMs’ ability to generate and classify ASCII text using PyTorch, Pix2struct, and CLIP, resulting in a comprehensive benchmark for model selection.

• Boosted model accuracy by 10% through web scraping and EDA, enhancing LLM benchmarking.

• Fine-tuned hyperparameters for CLIP and Pix2struct models, improving training efficiency and performance by 10%.

• Co-authored a paper on predictive modeling and forecasting accepted at AAAI’s Workshop on AI for Urban Planning.

• Engineered a neural network integrating spatial and temporal features, reducing NYC subway prediction errors to 0.42%

• Enhanced model performance by 12% through hyperparameter tuning and loss function optimization, achieving a significant reduction in validation loss with Keras.