projects I've worked on and currently building, from research to experiments
Currently building a multimodal reinforcement learning agent that learns to organize desktop folders using visual understanding and natural language instructions. Combines computer vision to interpret desktop screenshots with text-based commands to perform file organization actions.
Fine-tuned BERT model using K-Means clustering and sentence transformers to classify disaster-related tweets, achieving 90.49% accuracy through refined label generation on the Kaggle disaster tweets dataset.
Real-time emergency medical services (EMS) AI system using Knowledge Graphs to diagnose caller medical conditions faster than call duration. Integrates Whisper transcription, NER with spaCy/medspaCy, and Neo4j knowledge graphs to extract symptoms and predict health conditions, achieving 62.49% average processing time relative to call duration.
