Developing a tumor detection model using UNet and SAM2 with LLM, effectively segmenting tumors from real CT scans despite a small dataset (570 images)
Engineering an image segmentation pipeline employing UNet, SAM and SAM2 for automated tumor detection with accuracy score up to 90%
IT Support Assistant (Automation Focus)
City University of New York
Feb 2024 – Present
Automated task management and license compliance, while deploying and troubleshooting Windows images using MDT & SCCM
Managed and maintained Active Directory for 520+ users while optimizing IT operations with SysAid for incidents, assets, and service desk management
Co-Founder
BirdLens (Bird Species Identifier)
Jan 2021 – June 2021
Launched an AI-based web application to identify bird species in real-time using custom-trained TensorFlow models
Led a team of 5 engineers; directed MVP development, product strategy, and usability testing with local birdwatching communities
Achieved 97% detection accuracy and grew a beta user base to 50+ active users
Projects
RAG Document Q&A Chatbot
PythonRAGLLMLangChainFAISSStreamlit
Implemented RAG pipeline using NVIDIA NIM and LangChain, achieving fast and contextually accurate responses from PDF documents. Optimized document retrieval by vector embedding text with NVIDIA Embeddings, reducing query response time to milliseconds.
Engineered recommendation engine using Cosine Similarity based on Natural Language processing with 97% accuracy score. Implemented real-time web scraping with BeautifulSoup4 to extract IMDb reviews and performed sentiment analysis.
Achieved prediction accuracy with Quadratic Weighted Kappa (QWK) score of at least 0.80. Fine-tuned Funnel Transformer with 10-fold cross-validation, keeping validation loss under 0.40 with <5% error across 6 categories.
Leveraged computer vision through OpenCV with real-time facial recognition to automate attendance recording, reducing error by 98%. Utilized advanced facial recognition technology, mapping 128 features for highly accurate identification.
Led a team of 3 to streamline entrance exam procedures through a paperless, online system, enhancing transparency and efficiency. Demonstrated dedication to modern web development and database practices.
