Personal Website

Kenneth Zhang

Software engineer and ML practitioner working on RAG systems, text-to-SQL, and multimodal pipelines. I like building things that sit between research and production — tools that people actually use — and I occasionally ramble about movies, books, and music.

Hsinchu / Taipei, Taiwan

Profile picture

Work

Internships, research, and teaching roles

AI R&D Intern · LargitData

2025 – Present

Taipei, Taiwan

Building production AI systems: RAG chatbots with OCR/ASR, text-to-SQL BI assistants with auto charting, and MCP-integrated AI agents.

RAGLLMsDjangoPostgreSQLMCP

Research Assistant · CGV & MIS Lab, NTHU

2024 – Present

Hsinchu, Taiwan

Working on multimodal sports analytics: player search from audio + faces and transformer-based sports highlight detection.

Computer VisionWhisperTransformersSports Analytics

Research Assistant · AINS Lab, NTHU

2025

Hsinchu, Taiwan

Delivered LLM fine-tuning workshops and co-authored work on temporal correlation in large vision-language models.

LLMsFine-tuningVLMs

Teaching Assistant · Intro to Programming, NTHU

2025 – Present

Hsinchu, Taiwan

Designed projects, supported lectures, and helped students build solid programming fundamentals.

TeachingPythonCS Fundamentals

Projects

Selected things I've built or shipped

QubicX – Multimodal AI Assistant

2025

Desktop-like assistant that manages knowledge bases with RAG, plus OCR and ASR pipelines for documents, screenshots, and audio.

RAGLlamaIndexDjangoPostgreSQLWhisper

Wisbi – ERP AI Assistant

2025

Text-to-SQL assistant for ERP systems, combining RAG feedback loops with automated chart generation for self-service BI.

Text-to-SQLDjangoPostgreSQLLLMs

Detect AI-Generated Text

2024

Ensemble transformer classifier to detect AI-generated text, reaching 97.6% on a Kaggle benchmark.

TransformersPyTorchPEFT

Virtual Try-On App

2024

Mobile virtual try-on experience using image stitching, built with Flutter and a cloud backend.

FlutterFirestoreGoogle Cloud

Reviews

Book — The Pragmatic Programmer

2025

Notes on craftsmanship, communication, and practical heuristics that aged surprisingly well.

BookCraft

Film — Poor Things

2024

Wild, maximalist, and tender. Design language and score are a playground.

FilmDesign

Album — boygenius: the record

2023

Rich harmonies with quietly devastating lyrics. On repeat while coding.

AlbumIndie

Media

Talk — Building Practical RAG Systems

2025

An opinionated overview of retrieval, chunking strategies, and evaluation with LLM-as-judge.

TalkRAG

Post — Text-to-SQL with Feedback Loops

2025

From schema linking to guardrails: turning one-off queries into reliable assistants.

PostText-to-SQL

Demo — Multimodal Notes Inbox

2024

OCR + ASR pipeline to turn screenshots and voice memos into searchable notes.

DemoMultimodal