Liu Teng Yu

Developed a clinical consultation simulator for medical students and learner analytics, supporting voice-driven patient interviews with closed-loop instructor and learner tooling. Implemented authentication and learner profiles (e.g., Entra ID / Cognito) and built dashboards including deterministic LLM summaries for consistent outputs. Built an Entra attendance automation Playwright job to sign in, export daily attendance, validate CSVs, and upload to AWS accounts; added overlap locking, failure alarms, and an incident rerun strategy. Diagnosed intermittent production issues including upstream outages and added production-grade resilience.

Built an AI voice-cloning web app for recording upload, voice training, and generating lecture-style content or speech-to-speech outputs. Added long-form chunking, loudness matching across fades, and sentence-level review/repair flows; improved medical pronunciation accuracy with a pre-check tool, automatic speech checks, and duplicate-word detection. Created a reusable voice-profile API so downstream apps can request speech using only a profile ID and text. Also designed GPU infrastructure with autoscaling, distributed queueing, atomic handling, and pre-warming; load-tested up to 100–150 concurrent users and debugged scale-out root causes. Built real-time voice chatbots using WebSockets and phrase-by-phrase responses with users switching between fast and higher-accuracy engines; shipped an on-stage “AI co-host” site and a prompt editor so staff could safely change behavior without releases.

Digital Learning team. Built an AI voice cloning platform on AWS (GPU autoscaling, realtime voice chatbots), a clinical AI patient chatbot, Playwright automations on ECS Fargate, a WhatsApp AI auto-reply agent and React/Lambda features for a research platform.