At Microsoft Corporation, I designed and implemented an end-to-end GenAI assistant for clinicians, combining Azure OpenAI with agentic features. I built its multi-agent architecture and hybrid retrieval pipeline using FAISS and Neo4j.
I also implemented Model Context Protocol servers that connected FHIR APIs, CosmosDB, and Neo4j for secure healthcare data access. The assistant supports speech and image inputs, and I fine-tuned task-specific models for summarization, lab explanations, and symptom-to-condition mapping.
At Nike, Inc., I built demand forecasting and real-time recommendation systems, including a visual product similarity search pipeline that achieved sub-100ms search latency and improved CTR through A/B testing. Earlier, at Urbint, I developed AI systems for safety risk prediction and infrastructure damage detection; at Amazon, I built a Python tool that reduced manual product-listing QA by 70%.
