Yi-Wen (Claire) Liu

The firmware team managed 15+ parallel development machines with build times ranging from 2 to 60+ minutes, with no systematic way to detect anomalies or allocate hardware resources — decisions were made ad hoc.

Design and deploy an automated data collection and anomaly detection system from scratch, with zero dedicated infrastructure budget.

Built an end-to-end data pipeline using Python and Windows Batch scripting embedded directly into each machine's codebase (zero-touch ingestion). Designed a two-layer Google Sheets architecture separating raw ingestion from a management dashboard. Implemented a dynamic anomaly threshold algorithm (μ + 1.5σ) filtering ~70% of background noise, with real-time API uploads for live alerting.

Cross-regional firmware issues between Taiwan R&D and China QA teams frequently stalled device bring-up, with no structured triage process — bugs could live in local firmware, IBV logic, or vendor tooling.

Act as technical liaison to isolate root causes and establish reproducible validation workflows to accelerate bug resolution across three stakeholder groups.

Investigated ACPI/ASL-level anomalies by collaborating with IBV engineers to map fault boundaries. Built structured validation checklists where none existed, and documented triage decision trees to enable QA self-resolution.

Hybrid time-series forecasting, combining statistical models (ARIMA, SMA) with deep learning (LSTM) to reduce prediction error by up to 22% across financial, environmental, and public health datasets.

Built end-to-end data preprocessing workflows, ran forecasting experiments across statistical and deep learning approaches, and compared model behaviour across datasets to evaluate prediction performance.

A scheduling-based reservation system for allocating manpower and resources for the following month.