• Building state-of-the-art conversational AI agents that deliver accurate, scalable, and production-ready experiences.

• Designed and implemented a multi-RAG system within the LangChain agentic framework to streamline clinical document corpus management. Developed an AI-powered nurse assistant, significantly improving clinical review efficiency.

• Led experimentation on advanced LLM architectures, including multi-modal models, to enhance clinical document classification. Applied Parameter-Efficient Fine-Tuning (PEFT) techniques, achieving substantial performance gains across multiple downstream tasks.

• Developed scalable clinical automation pipelines, consolidating cross-team efforts to establish a standardized, extensible ML pipeline. Maintained critical ML infrastructure to support end-to-end deployment and handle increasing automation demands efficiently.

• Developed and deployed advanced heuristic models that directly prevented over $253 million in potential BEC (Business Email Compromise) losses for customers in 2021.

• Awarded U.S. Patent 11050698, “Message processing system with business email compromise detection,” June 29, 2021.

• Engineered and maintained neural network models (including CNN and bidirectional LSTM) to classify URLs, playing a pivotal role in blocking 40 million malicious phishing campaigns across 10 billion messages in 2021.

• Collaborated with cross-functional teams to implement random forest models leveraging TF-IDF and other features, effectively detecting various types of malicious attachments (e.g., XLSM, HTML, VBA).

• Designed and built an in-house brand logo detector using transfer learning, replacing the legacy OCR-based approach. This innovation significantly improved the identification of over 300 brand logos and enhanced overall phishing detection accuracy.