●Architected and deployed Virtual Lab, a physician-facing AI platform with FastAPI, real-time streaming, and enterprise SSO, enabling continuous automated document extraction and interactive multi-agent clinical decision workflows.

● Developed a ReAct-based multi-agent orchestration engine with MCP servers and A2A delegation, reducing manual clinical review time by 60%.

● Engineered a GPT-2 pipeline using PyTorch hooks to generate 768-dim patient embeddings from longitudinal notes, enabling EFI-based uncertainty estimation for AI-driven relapse prediction

● Built an end-to-end Databricks RAG pipeline ingesting 3,000+ clinical PDFs, enabling scalable enterprise knowledge retrieval.

● Optimized Databricks ingestion and embedding pipelines, cutting processing time from 5 days to 20 hours and enabling continuous dataset expansion.

● Implemented query rewriting and chunk reranking, improving answer relevancy by 60% and signal precision by 40%.

● Led technical coaching and mentorship for a 40-student, project-based software engineering course, teaching core concepts including relational data modeling, RESTful API design, client–server architecture, and MVC patterns

● Guided students through full-stack development using React, Flask, SQL, and ORMs, emphasizing clean architecture, separation of concerns, and maintainable code

● Facilitated office hours and 1:1 debugging sessions focused on query optimization, state management, backend logic, and frontend–backend integration

● Evaluated and graded coursework while delivering actionable feedback on code structure, performance, readability, and correctness

● Developed AI-driven models using K-means clustering and Latent Dirichlet Allocation (LDA) to analyze over 15 datasets, improving the accuracy of admissions decisions by 20% and reducing manual review time by 35%.

● Built an integrity check system to detect duplicate applications, reducing fraudulent entries by 25% and saving 100 hours annually.

● Optimized SQL queries for large-scale data extraction and analysis, leading to a 15% increase in processing efficiency and offering data-driven recommendations to streamline the admissions workflow.