<developer /> <researcher /> <innovator />

Building the future of
Human-AI Interaction

從物理研究到 NLP 應用,從全端開發到人機協作研究。專注於用 AI 解決真實世界的複雜問題, 創造有意義的技術產品。

5

國際科展一等獎

3

Production Projects

2

Research Publications

Possibilities

NLP Transformers React Native Multi-Agent Systems Full-Stack HCI Research

Featured Projects

Technical deep-dives into real-world solutions

隱途計畫 AI Documentation System

Role: CTO & Tech Lead

Production

為社工機構打造的 AI 輔助個案記錄系統。透過 LINE Bot + NLP 技術,將數小時的記錄工作壓縮到 10-15 分鐘, 已與更生少年關懷協會、富邦基金會等多個單位實地驗證。

System Architecture

Frontend Layer
LINE Bot
Conversational UI
Quick Replies
Processing Layer
NLP Engine
Intent Classification
Entity Extraction
Output Layer
Document Generation
Email API
Google Docs Export

Technical Implementation

Conversational Flow

dialogue_manager.py
def structured_interview():
    questions = [
        "訪視基本資料",
        "案主現況與觀察",
        "生活與支持系統",
        "介入內容",
        "風險與安全"
    ]
    
    for q in questions:
        response = collect_response(q)
        extract_entities(response)
        classify_intent(response)

NLP Processing Pipeline

nlp_processor.py
class DocumentProcessor:
    def process(self, text):
        # 自動標籤分類
        tags = self.classify_tags(text)
        
        # 實體萃取
        entities = self.extract_entities(text)
        
        # 生成結構化報告
        report = self.generate_report(
            tags, entities
        )
        return report

Impact & Performance

10-15min

完成時間

vs. 數小時

8

自動分類欄位

結構化輸出

9+

深度訪談

User Research

100%

實地驗證

Production Ready

LINE Messaging API Python NLP Intent Classification Entity Recognition Document Generation Google Docs API

OpenEvocracy Democracy Platform

Role: CTO | Partnership with Swedish NGO

International

為瑞典非營利組織開發的民主促進平台。使用 NLP 技術分析與整合公民意見, 將多元、甚至矛盾的觀點轉化為結構化對話,促進更透明、包容的民主參與。

NLP Processing Pipeline

Input
Raw Opinions
Analysis
Classification
Output
Structured Data

Core NLP Features

01

Opinion Analysis System

自動分類議題類別、情緒傾向、論點結構。使用 transformer-based models 進行多標籤分類。

classify("opinion_text")
→ {
    topic: ["environment", "policy"],
    sentiment: "positive",
    intensity: 0.87
  }
02

Semantic Similarity Engine

使用 sentence embeddings 計算語意相似度,識別不同表述下的相同訴求,避免重複計算。

similarity(text1, text2)
→ cosine_sim = 0.94
→ cluster: "same_concern"
03

Argument Structure Extraction

從自由文本中萃取核心論點、支持證據、反對意見,構建論證樹狀結構。

extract_structure(text)
→ {claim, support, counter}
04

Multilingual Support

支援多語言輸入與分析,確保民主參與的包容性。使用 mBERT/XLM-R 等多語言模型。

languages: ["en", "sv", "zh", ...]
Transformers BERT Sentence Embeddings Topic Modeling Multilingual NLP Semantic Analysis

LucidBook Mindfulness App

Role: Full-Stack Engineer

Production

正念冥想練習平台。從 0 到 1 獨立完成全端開發,包含用戶系統、練習追蹤、成就系統等完整功能模組。 重點優化 API 效能與 React Native 渲染效能。

Full-Stack Architecture

Frontend Stack

React Native + Expo 0.72.x
React Navigation 6.x
State Management Context API
UI Components Custom

Backend Stack

PHP 7.4+
RESTful API JSON
MySQL 8.0
Authentication JWT

Performance Optimization

-60% API

API Call Reduction

智能快取策略 + 請求批次處理

useMemo + useCallback
React.memo optimization
-60% Render

Re-render Reduction

優化 State 管理 + 組件拆分

Context splitting
Virtualized lists
100% Offline

Offline Capability

優雅降級 + 離線快取

AsyncStorage cache
Graceful degradation
React Native Expo PHP MySQL JWT Performance Tuning

AI Translation System for Construction

Role: CTO & NLP Research Lead

Research

建築業跨語言智能翻譯系統。不只做語意翻譯,更處理語氣、情緒與文化差異。 使用 Hugging Face Transformers 微調 MarianMT/mBART/M2M-100,針對工地溝通場景優化。

Model Training Pipeline

Step 1
Data
Step 2
Clean
Step 3
Train
Step 4
Evaluate
Step 5
Deploy

Training Configuration & Data Strategy

training_config.py
from transformers import MarianMTModel

# Model Selection
models = {
    'lightweight': 'Helsinki-NLP/opus-mt-zh-vi',
    'multilingual': 'facebook/mbart-large-50',
    'direct': 'facebook/m2m100_418M'
}

# Training Hyperparameters
config = {
    'epochs': 5,
    'learning_rate': 2e-5,
    'batch_size': 8,
    'warmup_steps': 500,
    'max_length': 128
}
data_pipeline.py
# Data Requirements
data_scale = {
    'minimum': 5000,    # sentence pairs
    'recommended': 20000,
    'optimal': 50000
}

# Data Sources
sources = [
    'internal_docs',
    'opus_corpus',
    'tatoeba',
    'back_translation'  # 回譯增強
]

# Preprocessing Pipeline
pipeline = [
    deduplication,
    length_filter,
    language_detection,
    tokenization
]

Model Comparison & Selection

Model Use Case Advantages Speed
MarianMT 歐洲語言、英中翻譯 輕量、速度快、效果穩定 Fast
mBART 多語言、低資源語言 支援 50+ 語言、泛化能力強 Medium
M2M-100 多對多翻譯 無需英文中繼、直接翻譯 Slow
Hugging Face Transformers MarianMT mBART M2M-100 Fine-tuning PyTorch

Human-AI Collaboration Platform

Role: CTO & Principal Researcher | Published at HCI 2025

Research

跨領域 AI 協作設計平台研究。首次系統性結合 LLMs、Multi-Agent Systems 與 Double Diamond 設計模型, 探索 AI 如何增強而非取代人類創造力。

📄 Paper: "Facilitating Interdisciplinary Co-Design with Human-AI Multi-Agent Collaboration"

Multi-Agent System Architecture

🌍
Environment
Agent
🧠
Psychosocial
Agent
Technology
Agent
👤
Human
Designer
⬍ Collaborative Dialogue ⬍
Design Cards Knowledge Base
Shared Context | Multi-Domain Insights | Structured Information

Double Diamond Integration

01 Discover
Information Gathering
Cards as collection of related information
02 Define
Insight Synthesis
Refine cards to consolidate understandings
03 Develop
Idea Generation
Cards as multi-knowledge base to prompt ideas
04 Deliver
Communication
Using cards to explain and discuss ideas

Research Contributions & Results

Enhanced Understanding

參與者表示 AI 使跨領域觀點更容易理解與整合

"The design cards made it so much easier to see how different perspectives could come together."
Creative Enhancement

AI 提供的建議激發了新的思考方向

"It really pushed me to think outside the box... It wasn't just suggestions—it was connections."
Efficiency Gains

AI 處理複雜資訊,讓人類專注於高層決策

"The AI was doing the heavy lifting with data, which freed me up to focus on the bigger picture."
LLMs Multi-Agent Systems Human-in-the-Loop Design Thinking HCI Research Qualitative Analysis

About Me

From physics to AI, from research to engineering

Background

從自然教育中成長,在物理研究中獲得 5 個國際科展一等獎。 從小說創作到程式開發,從哲學思辨到 AI 研究——這些看似分散的經歷, 其實都在探索同一個問題:如何用創造力理解並改變世界。

目前就讀清華大學電資學院學士班,主修人工智慧學程。 專注於自然語言處理、人機協作研究,以及如何用技術服務真實的社會需求。

Core Skills

NLP & AI

Transformers Hugging Face Fine-tuning Multi-Agent

Full-Stack Development

React Native PHP MySQL RESTful API

Research & Innovation

HCI User Research Experimental Design Academic Writing

Milestones

2025

HCI 2025 Publication

Human-AI Collaboration Research

Istanbul Youth Summit Best Group Award

2024

International Student Conference 70

AI Research Best Award

2023

Young飛行動計畫 & 清華創業車庫

Social Innovation Projects

2022

5× International Science Fair 1st Place

MOSTRATEC, KSEF, Step into the Future, FISSION, IS²E²R

Get In Touch

Let's discuss NLP, AI, or interdisciplinary collaboration

CURRENT STATUS

清華大學電資學院學士班
主修人工智慧學程
Based in Hsinchu, Taiwan