IBM-Singapore, Supply chain AP team. Data Scientist (Intern). 2018.8 - 2019.4

Use IBM-Watson to develop the chatbot which focus on specific manufacture domain. 1st stage, use the historical data to predict the next month average cycle time. 2st stage, use NLP, word2vec, QANet, LSTM to train and test the context cognitive capability which are able to distinguish the context changing and reset the dialogue automatically.

Coupang, GO-HLM Team, Software Engineer. 2016.6 - 2018.1

Coupang is one of the largest and fastest growing eCommerce platform, ranked the 1st place in Korean eCommerce industry. In 2016, received US$ 1 billion investment from Softbank and was included in 50 Smartest Companies by MIT Technology Review.

Our team and I focus on the logistics related business especially the Fleet Management Optimization (FMO) and Hub Sortation Management. Our systems provide system service for 9 hubs, 40+ camps, 300+ linehaul truck orders and over 400,000 parcels perday. 3 major projects I involved listed below.

(1) Hub Management Mobile Application

✼ Using React Native, Redux as the core mobile technique, combined with Android native modules, programmed the basic mobile components and whole application.

✼ Implemented the business features for Hub workers and stakeholders including administration permissions, parcels and trucks info scanning, status track of parcels and trucks, CRUD operations in listview.

✼ Utilizde the lightweight local database Realm to develop the app offline mode, which was leveraged to guarantee the data completeness while internet condition was unstable in workplaces. Depending on EventBus, recorded and uploaded the operation logs asynchronously and automatically to store in MongoDB.

✼ Developed version management platform for both server side and client side, used 2 strategies which were backend deployed updated package and CodePush hotfix.

(2) Truck Monitoring and Visualization System

✼ This monitoring system using techniques including Java, Spring MVC (internal MicroService), RabbitMQ, Angularjs and Google map Javascript api

✼ Architected spring batch job and implemented visualization of the realtime linehaul trucks locations and current route path in the Map to monitor individual status.

✼ Developed APIs to interacted by 3rd service companies Korean YLP (Korean location service) and Naver (navigation service). Got remain distance and time of each truck to calculate and analyze the rate of punctuality of CPT (Criteria Pull Time) and ExAT (Expected Arrived Time) in each route.

✼ Integrated with Slack and SMS to realize the functionality of sending notification to the operators belonging to each truck vendor.

✼ Developed the distance judgement algorithm to automatically toggle the truck status altering. Generated the statistics reports and graphs in Echart (daily and weekly) to estimate the performance of drivers, vendors and routes.

✼ This Monitoring system deployed in production, everyday recorded and provided services for 5 main truck vendors, 80+ routes and over 250 drivers.

(3) Loading Plan system (LPS)

✼ Implemented the simulation system using Java and Angularjs in backend and frontend.

✼ Designed and optimized the optimal truck dispatch algorithm to generate the most saving-cost dispatch of different vendors for each route perday based on historical data. Aggregated the total cost/saving and displayed in FMO dashboard.

✼ Aimed at relieving the pressure of intensive paths, the stopover dispatch method was developed to substitue the previous rough management to enhance the delivery efficency amongst the neighbouring camps.

✼ Based on the similarity of weekly dataset, regression algorithms were applied to predicate the hourly shipments. The optimal dispatch would be recommanded and automatically made the truck orders reservations.

✼ LPS project would save maximum ₩3000,000 transportation fee per week theoretically.

Liulishuo (英语流利说), Big Data Intern, 2015.6 - 2015.10

Liulishuo (NSDQ. LAIX) is the world's leading AI + Education company, was also selected in top 100 most promising AI companies in the world by CB Insights. The Company’s flagship mobile app product "Liulishuo" has more than 50 million registered users from 379 cities in China and over 175 countries around the world. Liulishuo has closed Series C funding which was led by China Media Capital and Wu Capital. Trustbridge Partners, IDG Capital, GGV Capital, Cherubic Ventures and Hearst Ventures also participated in this round. My internship mainly included in 3 parts.

(1) Liulishuo Score Data Visualization

✼ Aimed at validating the algorithmic robustness, constructed the visualization platform to display the batch speech scoring distribution of each language and algorithm versions, by using Apache Zeppelin.

✼ Integrated with current Spark clusters, using z-manager to operate the deployment.

✼ Utilized Angularjs (two-way binding) to realize the frontend layout.

✼ This job was folked by official ZEPL as the application exploration in showcase.

(2) Text-Audio Serialization and Scoring pipeline.

✼ Developed linguistic data serialization, with data scientists implemented linguistic scoring in both stream and batch modes in Scala.

✼ Applied Spark and cluster to realize the integration of scoring pipeline and stored output scoring files in HDFS.

✼ Understood the Spark including Map-Reduce, cluster management, broadcast variables and Spark SQL.

✼ Developed downloading audio data bulkly from Qiniu in distributed model.

(3) Others

✼ Developed web spider to crawl the Korean phonetic symbol in website by Python, Proxy and BS4.

✼ Code cleaning and pptimization.

✼ Learnt knowledge and read papers of the Natural Language Processing and HMM.