Work Experience

1

Customer Success Engineer, MLOps & LLM
InfuseAI 工合股份有限公司, 2021 - 2024
  • InfuseAI Introduction: AI, ML and LLM platform provider for tailored ML workflows. I provide end-to-end solution for enterprise users.
  • Management: Lead 3 colleagues to complete customer requirements.
  • Partnered with 20+ clients (semiconductor, banking, etc.) for tech requirements. Boosted revenue by 10%.
  • Solution Architect: Achieve 5x faster ML development via:Tools: Kubernetes, Docker, ML/Deep Learning/LLM model, RAG, AWS/GCPCloud, and open-source for MLOps and LLMOps solutions.
  • POC showcases for improved ML pipelines. Strategized model drift with SI partners.
  • Understood daily ML production; assisted customers in identifying operational pain points.
  • Technical Advocate:Engaged in AI, cloud, open source, and DevOps community days for tech updates.
  • Hosted online talks and articles on MLOps. In 2023, educated 100+ on MLOps technology.
  • Education: Trained 50+ users on product & MLOps for successful AI/ML projects.
  • Technical Docs: Authored 100+ articles aiding users & SI partners with k8s,AI/ML, & MLOps.
  • Operation Role: Collaborated with engineers, analysts, and product managers for customer system maintenance.

2

Senior Software Engineer, Transformation Office
FarEasTone Telecom 遠傳電信股份有限公司, 2018 - 2021
  • FarEasTone Telecommunications: A premier telecom service provider.
  • As a FarEasTone MA Program member, I spearheaded AI, data, and IoT-driven digital transformation.
  • Developed a machine learning service to produce fraud reports, slashing fraud detection by 90%. This initiative, rooted in deep telecom domain analysis, gained acclaim from 165 anti-fraud hotlines.
  • Utilized IoT sensor data to monitor room conditions, leading to the creation of data ETL, ML, CI/CD pipelines, and ML Monitoring. These methodologies later fueled our energy management projects.
  • Collaborated to employ cameras for traffic violation detection for the government, leveraging the YOLO-based model, ML pipeline, and edge computing. This solution was implemented in local Taiwanese government projects, meeting all stipulated requirements.
Personal Skill

1

ML/DL/LLM Framework
  1. Machine Learning: Support Vector Machine (SVM), Extreme Gradient Boosting Machine (XGBoost), Lightweight Gradient Boosting Machine (LightGBM), Decision Tree (Tree), Regression (Regression).
  1. Deep Learning: TensorFlow(Main), Keras, Pytorch.
  1. Image Model: OpenCV, YOLO-base Model.
  1. Model as a Service: TF-Serving, OpenVINO.
  1. Model Version Control: Machine Learning Process (MLflow).
  1. Programming: Python, object-oriented programming, FastAPI, Flask, Streamlit, etc.
  1. LLM:
  1. API: ChatGPT, OpenAI API, Google Gemini, Claude, Replicates
  1. Open Source: Llama-based, Mistral-based, Phi-2, zh-TW LLM Models
  1. Others: RAG, Fine-tuned LLM Model

2

DevOps and MLOps Concept
  1. Containerization development: Docker, Dockerfile, Kubernetes.
  1. Cloud platform: AWS, GCP, Azure Cloud.
  1. Deployment tool: Ansible.
  1. Version control: Git, GitLab.
  1. Monitoring: Elasticsearch, Prometheus, Grafana.
  1. CI/CD process: code review, code quality, container scanning, Pytest, Pylint.
  1. Agile development: Scrum, Kanban.

3

Product Management
  1. Customer introduction work - Presales: When contacting new customers, I will introduce them to let customers know how the company's products can make them successful.
  1. Customer demand architecture planning: Because the production of artificial intelligence is complex and requires planning first, and the concept of MLOps that I believe in is basically to standardize, streamline and automate the process of making AI, so that customers can It is handled in this way, so this part will be a bit like helping customers draw design drawings similar to the decoration industry, so that customers or third-party partners can construct according to the drawings.
  1. Customer POC: Because customers want to know that this can be successful, they will bring the topic back after discussing the topic with the customer, and use their own capabilities and the capabilities of the current LLM model to help the customer complete the POC sample he needs. Let them know that doing so will help the project generate value quickly.
  1. Product function planner: Because there are so many customer demands, it is impossible for the product to cover everything. Therefore, it is necessary to propose functional requirements to the company and conduct some PRD planning so that engineers can understand how to implement the functions. do.
  1. Communication with third-party SI partners: Because the product itself cannot be fully operated if it only relies on the company's status as the original manufacturer, so we must ask SI partners to assist with some customer needs, so there will be some problems in the process. Many of them need to communicate with third-party partners on technical and project requirements.
  1. Proposal report: When the customer is going to have a proposal report meeting for selecting a supplier today, I will produce the slides according to the previously planned content, so that on the day of the report, the customer can pay for the solution architecture we proposed.
Education

1

Industrial & Information Management, National Cheng Kung University (NCKU)
Master's Degree
  • Time: 2016 - 2018
  • GPA: 3.93 / 4.30
  • Thesis: The Development of Monotonic Support Vector Domain Description

2

Industrial Management, National Taiwan University of Science and Technology (NTUST)
Bachelor's Degree
  • Time: 2013 - 2016
  • GPA and Ranking: 3.89 / 4.00 (3 / 59)
  • Student Club: Transfer Student Association.
Made with Gamma