Sunnyvale, California, United States
• Built and deployed an AI auto draft feature in the internal portal to draft a response for new patient messages. 110,000+ drafts generated since deployment. Care team sent 40%+ AI drafts to patients without editing. Reduced the care team's time to respond to patient messages by ~54%, time per message median down from 1m22s to 38s. (Python FastAPI + OpenAI API + MongoDB)
• Built and deployed a patient summary feature in the internal portal that uses LLM to generate a two-sentence summary for any given patient, enabling the care team to get a patient overview within 10s. Implemented a caching mechanism using MySQL that balances token consumption, latency, and information recency. (Python FastAPI + OpenAI API + MySQL + MongoDB)
• Built and deployed an RAG chatbot in the internal portal to answer internal questions about any information in our enormous documentation with source citations. Used AWS Bedrock to chunk, embed, index, and retrieve document chunks. Answered 1,700+ questions since deployment. (Python FastAPI + OpenAI API + AWS Bedrock)
• Led a team of senior colleagues in the internal Hackathon. Designed and built 6 AI characters to remind and encourage patient logging in a Web/Mobile interface. Led the presentation and earned the highest score for presentation. (Python FastAPI + OpenAI API + HTML + JavaScript + CSS)
• Sampled 61,000+ patient chat messages and used LLM to classify each with one of 32 predefined patient intents. Analyzed intent distributions and provided insights into our chat automation effort. (Python + pandas + numpy)
