Muen Chiu邱沐恩

I.R.I.S. uses AI agents to investigate IT incidents and generate evidence-backed reports. It won first place at the 2026 TSMC IT CareerHack.

• Built with n8n, Google Vertex AI, context-aware RAG, and PostgreSQL.
• Designed a multi-agent workflow for root-cause analysis.
• Created a verifier loop to reduce hallucination by checking logs and private codebases.
• Projected 40% reduction in MTTR through automated documentation and knowledge reuse.

WaferCopilot helps engineers diagnose wafer defects using AI and simulate process adjustments before approval. Micron MIMORY Award finalist.

• Combined ResNet-18 wafer defect classification with Grad-CAM explainability and an RAG-powered LLM engine.
• Connected visual defect patterns with simulated FDC machine logs for traceable diagnostics.
• For simulatable defect cases, let engineers invoke a rule-based digital twin surrogate before acting on a parameter change.
• Reduced hallucination risk by grounding recommendations in SOPs and manufacturing signals.

In a simulation of Taipei’s Dunhua Main Line, reinforcement learning cut bus bunching volume by 22.9% versus fixed-headway dispatch. DAS 2026 Excellent Paper Award.

• Integrated TDX route data, Taipei Bus GPS APIs, and CODIS weather data into a dispatch simulation environment.
• Modeled departure control as a sequential decision problem and trained a PyTorch DDQN agent.
• Balanced bunching risk, passenger waiting time, and operating cost in the reward design.
• Reduced bunching volume by 22.9% and bunching occurrence time by 16.2% versus a fixed-headway baseline.

I used real H&M transaction data as market signals to design a generative AI framework that predicts fashion trends and visualizes potential products.

• Integrated LLaVA and Vicuna to produce trend-aware prompts.
• Used 106 weeks of H&M transaction data.
• Fine-tuned a Latent Diffusion Model with Direct Preference Optimization.
• Improved prediction accuracy by 4-7% over baseline, supporting inventory planning and market response.

I developed an AI pipeline that automates daily news highlight generation through summarization, speech synthesis, and image retrieval.

• Designed an automated news summarization pipeline.
• Combined NLP, text-to-speech, and image retrieval.
• Managed development milestones and module integration.
• Built daily news summarization workflow automation with Python and APIs.

I led a 6-member team to model a beverage chain's operations, identify bottlenecks, and evaluate process improvements through computer simulation.

• Led a 6-member team.
• Built a computer simulation to analyze operational bottlenecks and test improvements.
• Coordinated cross-functional discussions.
• Simulation results showed a 70% reduction in customer waiting time without additional operating cost.