October 2021 - Present

Senior AI & Backend Engineer  - Bitgin

• Build and maintain cloud systems for various products on the exchange:

  • Established Kubernetes (k8s) clusters from scratch on AWS and GCP using EKS and GKE.
    • Deployed foundational services such as Redis, Nats, InfluxDB, and product-related services using Helm. Configured and integrated RDS, implemented horizontal scaling for each pod, and automated scaling up when the load exceeded thresholds.
    • Monitored resources (CPU, memory, network) on CloudWatch, integrated Lambda for automatic notifications to Slack if resources surpassed thresholds. Configured vertical scaling, automatically scaling up when node resources were insufficient.
  • Configured Cloudflare for backend firewall and security protection, while also providing CDN for frontend performance enhancement.
  • Set up services like SNS, Cognito, and S3 for product functionalities.

• Established and maintained CI/CD processes:
  

  • Utilized GitHub Actions to create a CI/CD workflow to ensure the quality of deployments.
    

• Exchange Development:

  • Points System:
    • Designed and planned database tables for a points system, allocating platform points based on trading volume. Points could be used for other features. Collaborated with the frontend team to discuss GraphQL API and implemented it for deployment.

  • KYC Automatic Identity Verification:

    • Implemented KYC during user registration, integrated government z21 identity verification service RESTful API, and Jumio facial recognition service RESTful API. Designed database tables and automated the KYC process to reduce manual verification time for the customer service department.

• Strategy Product Development:

  • Workflow Framework:

    Designed a foundational workflow framework to accommodate diverse steps and parameters for each strategy. Planned SQL and NoSQL tables to facilitate the development of other strategies.

  • AI Research for Finance:

    • Conducting ongoing research on using deep learning to estimate future trends.

    • Generated various technical indicators and employed XGBoost, Random Forest for feature selection.

    • Explored various neural networks such as TCN, TFT, DLinear, FEDformer, Autoformer, TimesNet, PatchTST, Transformer, LLM, Time-LLM, etc..

    • Explored different papers and implemented key component such as Time2Vec, se-LSTM, GatedResidualNetwork, VariableSelection layers or blocks, etc..

    • Explored various connectivity methods and implementedlike Inception, Residual connect, SE blocks, etc..

    • Experimented with autoencoder, GAN, and different combinations.

    • Develop and fine-tune small version of LLM like GPT to training time-series data.

    • New data augmentation method for time series data

    • Goal:

      • short-term

        • AI indicator on Uniswap V3

          • Result: APR 300% ~ 500%

      • long-term (training with Reinforce Learning)

        • AI trader

• Introduced and maintained ISO 27001 information security standards:

  • Served as the highest-ranking executive on the operational side: Information Security Representative.
  • Adjusted internal development processes for information security compliance based on standards. Developed internal company guidelines, underwent external audits, and successfully obtained certification in a shorter timeframe than anticipated by consultants.
  • Passed the audit and obtained certification in the first year, and subsequently passed for two consecutive years.

  
  

  Skills: Go, Redis, PostgreSQL, NATs, PgBouncer, k8s, helm, GCP, AWS, Cloudflare, AI, Machine Learning, Deep Learning, Reinforce Learning, PyTorch, LLM, Transformer

May 2016 - October 2021

Senior Deep Learning Engineer - Touch Cloud

• Products:

  • Abaci/KekkAI:

    • Primarily focused on object recognition for people and vehicles using deep learning. Used for various functionalities such as object intrusion detection, object traffic counting, applied in settings like factories, violations, malls, security, etc. Customization of objects for clients and execution on edge devices.
    • Product Leader: Discussed with supervisors, designed and planned the product, led a team of 5 people.
    • Developed new tracking algorithms based on model outputs.
    • Developed a fence system for post-processing to achieve detection and counting.
    • Developed a framework to replace processing components corresponding to different models in the product system flow.
    • Used SQLite to plan tables for storing settings and results.
    • Deployed using Docker on edge devices.
    • Transferred trained models through OpenCV to OpenVINO for execution on Intel GPU for inference.

  • Grey Goose:

    • Collaborated with VMS foreign company NxWitness to integrate certain functions of the above product into NX's VMS as a plugin.
    • Product Leader: Discussed with supervisors, designed and planned the product, led a team of 2 people.
    • Assisted development team members in porting relevant functions to the plugin.

• Industry-Related Projects:

  Common process for the following projects: Collect image data and clean it, try various image pre-processing techniques such as distortion, rotation, flipping, etc.

  • Wafer Defect Classification and Detection:

    • Used InceptionV4 for initial classification training. After achieving a certain accuracy, replaced the head with SSD and used transfer learning to train object detection.

  • Gas Meter Reading Detection:

    • Used MobileNetV3, replaced the head with SSD to detect specific number areas. In the subsequent process, cropped the images of number areas. After collecting a certain amount of images, fine-tuned using the CTC model to achieve gas meter reading resolution.

  • PCB Defect Detection:

    • Segmented images from high resolution into numerous 512x512 small images. Adjusted the original bounding boxes as training data. Used InceptionV4 with the head replaced by SSD for training and conducted hyperparameter optimization.

• OpenStack Project

  • Building OpenStack cloud system
  • Planning building cloud system
  • Auto deploy with Ansible
  • Develop new network feature with Neutron

Skills: Golang, Python, Gin, Flask, Redis, SQLite, PostgreSQL, OpenCV , OpenVINO, Caffe, TensorFlow, Keras, Deep Learning/Machine Learning, Docker, Git, DevOps, CI/CD

March 2013 - May 2016

Research and Development Engineer - EstiNet Technologies Inc.

• E-Central Product

  Implement network device control, VLAN configuration, switch control, QoS control, and port traffic statistics using SDN.
  • Program architecture planning, design, and implementation.
• RRM Project
  Implementing an RRM (Radio Resource Management) module to control WiFi power, channels, and other radio-related resources.
  • Project Leader
  • Program architecture planning, design, and implementation.
  • Responsible for communicating with clients to understand requirements and planning project timelines.
  • Responsible for managing the development flow and task assignment.
• VoIP Project
  Implement packet routing and packet sniffing for VoIP (Voice over IP) using SDN.
  • Program architecture planning, design, and implementation.
  • Participation in exhibitions and demos.
• Bandwidth on Demand Project
  Implement bandwidth adjustment and usage billing between multinational companies using SDN
  • Program architecture planning, design, and implementation.
• C Model Project
  Implement a C Model to simulate the operation of the openflow switch ASIC chip, for comparison with the actual ASIC chip.

Skills: Java, C/C++, Docker, Linux, SDN, Network Protocols, OpenFlow