陈天佳 Email & Phone Number

• Improved the efficiency of system reinstallation and configuration for new customers to try OCP servers by using MAAS/Juju and Cobbler• Conducted function and pressure test for servers and clusters to prove its performance.• Participated in open compute server design with hardware company, such as Inspure and Quanta Inc.

中国 北京市区

• Detected cooling problems in data centre by using a workload-independent big data model then conserved data centre cooling power by smarter job scheduling. As a result, we have saved 10% of electrical power.• Proposed a state graph approach to debug and troubleshoot OpenStack problems.• Used reinforcement learning for the path selection in the data center network to reduce its latency.• Built entire data centre from the very beginning, explicitly designing air duct system, electrical power system, PDU, servers arrangement and monitoring system.• Configured and implemented entire cloud system, OpenStack/Docker, Harbor, Hadoop, Tensorflow, HDFS, NFS, Ceph, Swift, PXE, Cobbler, LDAP and distributed memory speed storage system Alluxio.• Through monitoring system to find out idle servers and let it mines Bitcoin to cover part of the electricity fee.

• Nowadays many online services are rife with fraud users and we proposed a novel distance metric over user profiles with multiple categorical features to integrate the subspace grouping characteristics of fraud users into the same high dimensional data space.• Built up scalable, multi-layer collision system to visualize the output of algorithmic fraud detection methods. This system has a friendly visualization interface for people without a programming background.

Built up an intelligent exhibition hall from the beginning. I deployed raspberry pi with infrared monitor and display device. When people step in the hall, the raspberry pi sensor will detect it and turn on all the devices, then people can scan the QR code on the wall and control all attractive devices developed by professors, and their theories will appear on the screen. All the information students want to know can be detected from the voice that students speak aloud, Raspberry Pi receives the voice and translates it to the command line just like Siri.Design and Maintain website.

北京丹棱街

• Collected data from office 365 users’ behavior, and used the FPG algorithm to pre-load tenants’ next behavior.• Trained tenant data using Smoke and MapReduce, and used Report Builder to design UI and visualise data.

北京中关村

同事合作管理微博网页平台，建立运营“微活动”和“微群”系列产品

Master'S Degree, Computer Science

学士, 信息与通信工程