# Jia Shi > Building things Location: San Francisco Bay Area, United States Profile: https://flows.cv/jiashi ## Work Experience ### Software Engineer @ OKX Jan 2024 – Present | San Jose, California, United States ### Software Engineer @ Sperta Jan 2021 – Jan 2024 | San Francisco Bay Area The Founding Engineer at Sperta ### Software Engineer II @ Uber Jan 2021 – Jan 2021 | Sunnyvale, California, United States Spender Risk - Chargeback frauds - Chargeback representment - Arrears - DNS risk ### Software Engineer II @ Uber Jan 2020 – Jan 2021 | Palo Alto, California, United States Doc verification and deduplication - Training flow to preprocess the document images. - Process thousands of types of document images on Uber platform. - Verify document images by comparing with the profile images. - Dedupe document images based on different important component from document images. ### Software Engineer @ Uber Jan 2019 – Jan 2020 | Palo Alto, California In house face verification system. - Use Computer Vision technique to identify the fake/duplicate images on accounts on Uber platform. - Large scale similarity comparison and clustering. - Leverage Apache Spark to make it scalable. ### Machine Learning Engineer @ University of Michigan Jan 2018 – Jan 2018 | Ann Arbor Environment: Python, Tensorflow, Git -Worked independently with a professor in EECS department on the project to replace traditional PID controllers on UAV with the controller based on reinforcement learning. -Searched suitable reinforcement learning algorithms for continuous control -Implemented a simulation and an exploration model for UAV to generate on-policy trajectories. -Programmed the Actor and the Critic models with Tensorflow to apply the Deep Deterministic Policy Gradient (DDPG) algorithm to teach the UAV to archive the goal position. -Modeled the movement of UAV from one waypoint to another as Rotate-Transit mode to avoid hitting obstacles. -Planned to apply more functionalities such as completing tasks with key objects recognition and localization, adjust the system to make the Critic mode ### Project of SLAM @ University of Michigan College of Engineering Jan 2018 – Jan 2018 | Ann Arbor, MI Project Environment: C++, SLAM, Raspberry Pi 3, Beaglebone Green -Set up and connected Beaglebone Green and Raspberry Pi 3. Synchronized the programs on Beaglebone Green which does not connect to a time server with programs on Pi by sending an lcm message of the current time on Pi every second. -Implemented Occupancy Grid to describe the space around the robot via the probability that each cell of the grid is free or occupied. -Created a particle filter for the SLAM system by designing and implementing action model and sensor model to utilize data from Lidar and odometry encoders. -Developed and programmed A* path planning algorithm and exploration rules to make the car be able to explore the mazes and solve robot kidnapping problems autonomously. -Communicated with professor and classmates from other teams on designing possible start strategies of robot kidnapping problem for solid performance of robot. ### Software Engineer Internship @ News Break Jan 2017 – Jan 2017 | San Francisco Bay Area Environment: Python, MongoDB, Java, Apache Kafka, Git Content Extraction Monitor: - Used titles from googlenews, flipboard and bing as reference and calculated levenshtein distance between extracted titles and reference titles to judge the correctness of title extraction. - Found duplicate shingles between different pages from the same website to find the common extraction errors in the same domain. Used content from topbuzz as reference and compute the percentage of difference between reference and extracted content by levenshtein distance. - Compared the public date from googlenews with extracted public time to determine the correctness of time extraction. - This monitor made the backend team know how many problems content extraction has and which website the problems are from. URL filter: - Parsed the crawled URLs by '/' and find the frequencies of different combinations of parsed URLs to find commonly appeared URL patterns which could be the patterns of non-articles' URLs. - This filter removed the non-article pages which could improve the performance of content extraction and increase the precision of correctness calculation. Content Extraction Importing tool: - Implemented a tool to import xpath or css-selector rules that are used to fix content extraction errors into database. Entertainment and Tech Knowledge Graph: - Built knowledge graphs to connect related terms in technique and entertainment area together by traversing themselves, their parents terms, and subsidiaries from Wikipedia, IMDB, billboard and so on. - These graphs facilitated the recommendation team work by giving them the relationship between terms in an area. ### Internship @ Beijing Benz Automotive Co., Ltd. Jan 2014 – Jan 2014 | Beijing City, China Worked in Research and Development Center: - Collected and analyzed data from the experiments for the prototype vehicles with different labs. - Tested functionalities of electric modules in -30℃, 0℃ and 30℃ condition to find out and fix fails. - Did trouble noises and vibration test by putting the prototype vehicle in both normal and extremely cold circumstance. - Examined if deformation happened on prototype vehicles and parts by setting many reference points on the body of vehicle and taking pictures before and after testing under extreme temperature and humidity condition. ## Education ### Master of Science - MS in Robotics University of Michigan ### Bachelor of Science - BS in Electrical and Electronics Engineering University of Illinois Urbana-Champaign ### Beijing No.4 High School ## Contact & Social - LinkedIn: https://linkedin.com/in/jia-shi-31655b155 --- Source: https://flows.cv/jiashi JSON Resume: https://flows.cv/jiashi/resume.json Last updated: 2026-04-05