I am Ko-tik Lee, a Machine Learning Engineer with over 3 years of experience in developing and deploying AI models. My expertise encompasses speech processing, computer vision, and large language models. Throughout my career, I have been focused on optimizing model performance, building scalable pipelines, and delivering cutting-edge AI solutions.

Expereince / Education / Awards / Publications / Personal Projects

Experience

• Taiwan AI Labs
  Machine Learning Enigneer | Aug. 2021 - Nov. 2024
  • Speech AI Enhancements:
    • Speaker: Led the development of state-of-the-art speaker diarization for Yating Transcriber, outperforming two key competitors in real-world applications and achieving top results on academic benchmarks. See my master’s thesis for more details.
    • ASR: Reduced Word Error Rate (WER) by 20% in specialized Automatic Speech Recognition models using self- supervised learning, providing transcription services to financial and medical sectors.
    • TTS & VC: Enabled rapid deployment of multilingual Text-to-Speech and Voice Cloning models for video translation and public broadcasting with minimal data (5 minutes) or zero-shot capabilities, enhancing content accessibility and reach.
    • Products: Yating Studio
  • Advanced NLP & LLM Applications:
    • RAG: Improved customer service chatbot user experience by engineering a Retrieval-Augmented Generation pipeline that enhanced retrieval accuracy by 25%, while maintaining a balance between accuracy and speed.
    • Multimodal LLMs: Architected a novel pipeline leveraging Multimodal LLMs, RAG, and custom modules for efficient summarization and retrieval from long-form video content (1hr+).
    • Products: Federated GPT
• Teaching Assistant at NTU
  Machine Learning (Fall 2021), Prof. Hung-yi Lee | Mar. 2021 - Jul. 2021
  • Designed two assignments on BERT and Explainable AI for a class of over 1,000 students.

Education

• National Taiwan University
  • M.S. in Electrical Engineering, EECS (Sep. 2022 - Jan. 2024)
    • Advisor: Hung-yi Lee, Speech Processing and Machine Learning Lab
  • B.S. in Electrical Engineering, EECS (Sep. 2018 - Jun. 2022)

Awards

• 1st Place in Cadence Data Structure and Programming Contest [Github]
  • Project in C++: FRAIG - Functionally Reduced And-Inverter Graph
• Presidential Award (Dean’s List, award to top 5% in class)

Publications

• Improved Speaker Diarization Based on Speech Foundation Models [Master’s Thesis]
• SUPERB: Speech processing Universal PERformance Benchmark [Interspeech 2021]

Personal Projects

• Eng Vocab Learner - Personalized English Learning Web Application [Github]
  • Final Project of Web Programming (Fall 2022)
  • A React-based vocabulary learning application with MongoDB integration, inspired by commercial language learning platforms. Features include:
    • Interactive flashcard system
    • Customizable learning progression
    • Adjustable difficulty levels
    • Progress tracking