Huang Chi-Wei

M.S. student in NCKU, passionate about AI computing optimization and computer architecture, specializing in model compression, quantization, and efficient LLM inference. Experienced in optimizing LLMs for resource-constrained hardware and developing advanced computing systems.

Research Focus: Model Compression, Quantization, Efficient LLM Inference
Advisor: Prof. Chai-Chi Tsai  
Lab: AI Systems Lab

Google Research Project – AI-Based Speech Screening Tool

• Integrated an LLM-based ASR model (Whisper) into a Flutter app, optimizing Whisper quantization for real-time speech analysis on Android (Pixel 6).

Capstone Project - Enhancing User Privacy Through Local Deployment of LLMs

• Conducted model workload analysis and applied state-of-the-art quantization techniques on Llama models, using MLX (Apple ML) for deployment on Apple Silicon (M-series) to improve efficiency and privacy.
• Developed a Next.js UI and built a FastAPI backend with Ollama, optimizing MLX server for real-time response streaming.

Course Project -  Performance Modeling and Optimization on μRISC-V Processor

• Implemented RISC-V assembly for array multiplication, analyzing execution time and CPI to classify CPU vs. Memory-bound tasks.
• Optimized array multiplication using RISC-V V Extension, achieving speedup >6x over baseline execution.

Capstone Project - Matrix Multiplication Accelerator

• Designed a systolic array accelerator in Chisel with weight-stationary and output-stationary dataflows, optimizing GEMM operations for low-latency AI inference.