30+ hrs
Efficiency Gain
Monthly manual workload reduced through AI agentic workflows.
Taipei, Taiwan
Shipped AI workflows across commerce, insurance, and banking
Reduced manual workload by 30+ hours per month, improved RAG reliability, and led AI risk prevention showcased to 300+ industry professionals.
STRATEGIC IMPACT
Bridging Intelligence with Real-World Impact
I ship AI systems that cut manual work and improve decision speed.
90%
Optimization
Reduction in LLM hallucinations via RAG and vector database architecture.
5+
AI Solutions
End-to-end AI bots developed across FinTech and E-commerce domains.
20B
Model Expertise
Parameter scale of fine-tuned LLMs using LoRA/QLoRA techniques.
300+
Public Engagement
Professional visitors engaged at the Taipei FinTech Expo.
Feature Stories
Fueling Enterprise Velocity with Operational AI
Enterprise Agentic AI Orchestration
Built production Agentic AI to automate enterprise monitoring.
Problem
- Fragmented cross-system monitoring workflows requiring high-frequency manual verification.
- Inconsistent data structures hindering automated reporting and archival processes.
Approach
- Defined requirements for Agentic AI bots utilizing cross-database natural language querying.
- Designed LLM-based ETL pipelines for automated structured information extraction and labeling.
- Orchestrated technical alignment between product and ML engineering teams for internal AI platforms.
Outcome
- Standardized AI deployment frameworks across core business and Public Relations units.
- Achieved significant cost-efficiency through automated multi-table data verification.
Problem
- Fragmented cross-system monitoring workflows requiring high-frequency manual verification.
- Inconsistent data structures hindering automated reporting and archival processes.
Approach
- Defined requirements for Agentic AI bots utilizing cross-database natural language querying.
- Designed LLM-based ETL pipelines for automated structured information extraction and labeling.
- Orchestrated technical alignment between product and ML engineering teams for internal AI platforms.
Outcome
- Standardized AI deployment frameworks across core business and Public Relations units.
- Achieved significant cost-efficiency through automated multi-table data verification.
Agentic AI
LLM ETL
Prompt Engineering
System Alignment
Compliant GenAI Systems Integration
Improved RAG precision for regulated insurance document workflows.
Problem
- High sensitivity to hallucination in regulated document search and retrieval tasks.
- Need for data-driven product prioritization within strict financial compliance frameworks.
Approach
- Optimized RAG retrieval paths by evaluating LLM performance within internal document systems.
- Leveraged machine learning algorithms to uncover high-correlation product insights for marketing.
- Managed metadata lineage tracking to ensure data integrity for regulatory reporting.
Outcome
- Enhanced response trust by reducing retrieval errors in document-heavy operations.
- Successfully integrated AI prioritization models with enterprise data governance standards.
Problem
- High sensitivity to hallucination in regulated document search and retrieval tasks.
- Need for data-driven product prioritization within strict financial compliance frameworks.
Approach
- Optimized RAG retrieval paths by evaluating LLM performance within internal document systems.
- Leveraged machine learning algorithms to uncover high-correlation product insights for marketing.
- Managed metadata lineage tracking to ensure data integrity for regulatory reporting.
Outcome
- Enhanced response trust by reducing retrieval errors in document-heavy operations.
- Successfully integrated AI prioritization models with enterprise data governance standards.
RAG Tuning
Data Lineage
Machine Learning
Strategic Prioritization
Full-Stack AI Service Orchestration
Unified financial analytics and geolocation into one AI flow.
Problem
- Disconnected financial tools causing high cognitive friction for analytics and branch search.
- Inaccurate vector retrieval leading to unreliable customer service responses.
Approach
- Architected a unified flow integrating low-code orchestration, vector databases, and real-time APIs.
- Implemented RAG optimization using local vector databases to ensure high-fidelity financial Q&A.
- Managed full-stack integration between React frontend and AI-driven backend services.
Outcome
- Delivered a seamless, interactive financial assistant with dynamic visualization and geolocation.
- Reduced LLM hallucinations by 90% through improved vector search and retrieval logic.
Problem
- Disconnected financial tools causing high cognitive friction for analytics and branch search.
- Inaccurate vector retrieval leading to unreliable customer service responses.
Approach
- Architected a unified flow integrating low-code orchestration, vector databases, and real-time APIs.
- Implemented RAG optimization using local vector databases to ensure high-fidelity financial Q&A.
- Managed full-stack integration between React frontend and AI-driven backend services.
Outcome
- Delivered a seamless, interactive financial assistant with dynamic visualization and geolocation.
- Reduced LLM hallucinations by 90% through improved vector search and retrieval logic.
AI Orchestration
Vector DB
React Architecture
API Integration
GitHub Side Projects
Applied AI builds across product, finance, and automation
5
Repositories
18
Focus Areas
Primary Build
View on GitHubCTBC_AIBankingApp
APP for AI Application of CTBC Bank.
Built an AI-oriented banking app prototype with clear task routing and maintainable TypeScript modules.
AI AppBanking UXTypeScriptFrontend Architecture
GPT_OSS-Finetuned
GitHubGPT_OSS 20B model finetuned with specific financial datasets.
Finetuned a 20B-class open model on finance corpora and structured repeatable evaluation workflows.
Fine-tuningFinancial NLPPrompt EvaluationExperiment Tracking
Financial-Mapping-Modal
GitHubAI-powered Financial transaction classifier.
Implemented a transaction mapping pipeline for cleaner downstream reporting and reconciliation.
Transaction ClassificationData MappingPythonFinance Ops
Yolo-v8-recipt-model
GitHubFinancial receipt recognition using YOLOv8.
Trained a YOLOv8 pipeline for receipt field recognition to reduce manual document handling.
Computer VisionYOLOv8OCR PipelineAutomation
Dailynews_Bot
GitHubDaily news Notification Bot.
Built a Python bot to aggregate and schedule daily news delivery with minimal manual effort.
Notification BotAutomationSchedulingPython
Case Studies
Production-grade AI delivered in high-stakes environments
Shopee AI Transformation
Architected Agentic AI workflows to automate enterprise data verification.
Problem
Enterprise monitoring and CSR reporting relied on fragmented, high-touch manual coordination.
Approach
Designed Agentic AI pipelines and LLM-based ETL for structured information extraction.
Outcome
Reclaimed 30+ manual hours per month through automated multi-system verification.
Agentic AI
LLM ETL
Workflow Automation
Mercuries GenAI Systems
Optimized RAG precision for compliant, document-heavy financial workflows.
Problem
Regulated insurance operations required high-fidelity AI responses and strict data lineage.
Approach
Integrated LLM solutions with RAG tuning and ML-driven product correlation analysis.
Outcome
Strengthened response reliability and reporting integrity for IFRS 17 frameworks.
GenAI
RAG Tuning
ML Analysis
IFRS 17
TMBA x CTBC AI Orchestration
Unified financial analytics and assistant flows into a seamless service.
Problem
Disconnected tools for analytics and Q&A created high cognitive friction for users.
Approach
Built a full-stack orchestration layer connecting vector search, charts, and real-time APIs.
Outcome
Reduced LLM hallucinations by 90% while providing one-stop financial insights.
React
n8n
Qdrant
Vector Search
Experience
Timeline
Roles across AI product delivery in finance and commerce.
Timeline
Nov 2025 - Present
AI Transformation Product Manager Intern @ Shopee
AI Transformation
- Architected end-to-end Agentic AI pipelines, including a Charity Management ChatBot and SOV Detection Bot.
- Engineered LLM-based structured data extraction (ETL) for automated enterprise monitoring.
- Orchestrated cross-functional alignment between product, data, and ML engineering teams for internal AI platforms.
Feb 2025 - Jul 2025
Information System Designer Intern @ Mercuries Life Insurance
Information Systems Design
- Integrated RAG-based Generative AI solutions into enterprise document management systems.
- Strategized AI application roadmaps using VDF methodology to prioritize organizational digital transformation.
- Applied machine learning algorithms to analyze insurance product correlations and enhance bundling logic.
Jul 2024 - Feb 2025
Anti-Money Laundering (AML) Project Intern @ Taishin International Commercial Bank
AML & AI Risk Prevention
- Spearheaded the public exhibition of AI-driven abnormal transaction detection models at the Taipei FinTech Expo.
- Managed high-stakes KYC and client monitoring workflows for corporate and individual banking services.
INTERESTS
Life Beyond the Terminal
Personal systems for staying curious, grounded, and active.
