AI Engineer | Medical ASR | Data & Workflow Systems
Turning Domain Know-how into Production-Grade AI Systems
Yu-Yuan Chang
Specialties: Medical ASR, Data Pipeline Engineering, Model Fine-tuning, Workflow-Oriented AI Deployment.
- Collaborating with 9 hospitals on medical ASR development
- Built Whisper / AdaLoRA / Multi-GPU training & real-time streaming architecture
- Managed NT$60B+ production scheduling & cross-department digital transformation
About Me
With over 8 years of experience in manufacturing and global operations, I previously served as a Production Planning Lead, managing high-stakes scheduling for operations exceeding NT$60 billion in annual revenue. I spearheaded key digital transformation initiatives, including VMI implementations and cross-departmental automation systems.
Driven by a passion for data-led impact, I earned an MSc in Data Analytics (Distinction) from Queen Mary University of London before transitioning into AI. Currently an AI Engineer at ITRI, I specialize in Medical ASR, overcoming clinical challenges like domain-specific terminology and acoustic noise through advanced model fine-tuning and data engineering.
I apply Lean Manufacturing principles and production scheduling logic to optimize AI training pipelines and resource allocation. Whether orchestrating factory floor logistics or deploying distributed training clusters, I bring analytical rigor to every stage of the workflow.
NT$60B+
Annual Revenue Managed Globally
9
Hospitals in ASR R&D Collaboration
8+ Yrs
Cross-Domain Industrial & AI Experience
9
Cross-Functional Team Members Led
Career Path
From global sales and production management to AI R&D—every step builds a unique perspective for solving complex problems.
2015 – 2019
International Sales Strategy
You-Ji Machine / Shanhua Industrial
Spearheaded global market expansion and client retention, mastering cross-cultural negotiation and B2B business development.
2020 – 2023
Production Planning Lead
Yieh Phui Enterprise
Managed a team of 9 to oversee NT$60B annual production scheduling. Led the development of VMI and automated scheduling systems.
2023 – 2024
MSc Data Analytics (Distinction)
Queen Mary University of London
Focused on machine learning, predictive modeling, and BI. Successfully transitioned from management to a technical core.
2025 – Present
AI Engineer
Industrial Technology Research Institute (ITRI)
Specializing in Medical ASR. Responsible for model fine-tuning, distributed training, and real-time streaming deployments across 9 hospitals.
AI Projects
Showcasing full-stack AI capabilities—from medical ASR research and RAG applications to intuitive frontend AI tools.
Medical ASR Fine-tuning
The Challenge
Standard ASR models often fail in clinical settings due to heavy acoustic noise and highly specialized medical terminology.
The Solution
Fine-tuned Whisper Large V3 Turbo using AdaLoRA; engineered a multi-GPU distributed training pipeline and a real-time WebSocket service.
The Impact
Achieved a significant reduction in Character Error Rate (CER) for medical terminology; deployed for real-time inference across multiple hospital sites.
Self-Refining Framework for ASR
The Challenge
The high cost and scarcity of annotated medical speech data limit the scalability and performance of ASR models.
The Solution
Developed an ASR ↔ TTS closed-loop self-refining system (inspired by MediaTek research) to leverage synthetic data for performance gains.
The Impact
Drastically reduced dependency on manual annotations via a 6-stage pipeline featuring 46 modular utility tools.
Corpus Annotation Pipeline
The Challenge
Raw audio from 9 different hospitals lacked standardization, making it unusable for direct model training.
The Solution
Built an end-to-end pipeline integrating Stable-Whisper forced alignment, Meta Denoiser, and hallucination detection.
The Impact
Produced high-quality, standardized ASR datasets with an automated, multi-engine processing workflow.
Recommender System Analysis
The Challenge
Addressing data sparsity and cold-start issues to identify optimal recommendation strategies for diverse user scenarios.
The Solution
Implemented and benchmarked 7+ algorithms (CF, SVD, ALS, MLP, Autoencoder) on the MovieLens 20M dataset.
The Impact
Delivered a comprehensive research report featuring cold-start simulations and visual performance benchmarks.
MedicalQA-RAG
The Challenge
Clinicians often struggle to quickly retrieve accurate information from vast, scattered medical documentation.
The Solution
Engineered a RAG-based Q&A system using LangChain, Qdrant vector database, and a multi-LLM backend architecture.
The Impact
Streamlined the transition from document retrieval to intelligent Q&A, including automated health education report generation.
AI Pipeline Designer & Image Enhancer
The Challenge
Non-technical users need an intuitive way to compose and execute complex image processing workflows.
The Solution
Built a visual pipeline designer using React 19 and TypeScript, integrating Gemini API for intelligent image optimization.
The Impact
Enabled drag-and-drop workflow customization, allowing AI to automate and optimize image processing steps.