I built an autonomous multi-agent personal assistant using LangGraph and specialized CrewAI agents. It automated six multi-step productivity tasks and used Firestore for long-term memory and context across user sessions.
For a deepfake detection system, I combined CNNs and Vision Transformers to reach 95.4% accuracy on held-out video data. Temporal-inconsistency preprocessing reduced the false-positive rate by about 14% against baseline models.
I also designed and trained a GAN that generated more than 5,000 synthetic X-ray and MRI images for downstream research. I deployed it behind a Streamlit interface for interactive evaluation and visualization.
During my Data Analyst externship at Beats by Dre (via Extern), I built an NLP model to classify more than 5,000 Amazon reviews by sentiment and theme. I also created web-scraping pipelines, analyzed the results with Python and data-science tools, and integrated Google Gemini into the research workflow.
