Innovatively engineered dynamic server-side configuration for client-side devices, enhancing flexibility and responsiveness. • Developed infrastructure within Apollo backend using Java and Kotlin.• Integrated dynamic config with Kindle iOS App, showcasing real-time behavior adjustments.• Achieved a 6% reduction in annual data processing costs by stopping the emission of 5.1% of total data that was deprecated, cutting waste and server load.

Ecological is a sustainable tech project group developing a system of apps to help consumers make more sustainable choices in their daily purchases. By providing users with sustainable purchase options, discount opportunities at partner stores, and personalized environmental impact analyses, we inspire and encourage eco-friendly decision-making.Responsibilities:• Building the backend of the sustainable lifestyle app to handle data processing and interaction using Express.js. • Leveraging React Native to create responsive and visually appealing UI for the mobile app, fostering user engagement. • Creating a robust REST APIs architecture with clear endpoints, data validation, and OAuth protocols, seamlessly interfacing with Square APIs, to retrieve transaction history data.• Deploying the server application in an AWS server using AWS Amplify and AWS Lambda frameworks.

Facial Expression Recognition (FER) is the task of classifying the expressions onfacial images into various categories, such as anger, fear, surprise, happiness, etc. FER is movingfrom laboratory-based theory demonstrations to practical real-world applications, but there are stillmany challenges to achieving a working system in the field. Automatic FER has typically facedtwo major challenges: lack of sufficient and/or demographically representative training data, andvariation in image quality parameters. Although there are 6 or 7 sentiments that are frequentlyidentified, in order to deliver a benefit to FLASH, this system only needs to categorize intosatisfied (good/neutral experience) or dissatisfied (poor/bad experience). The team will utilize ahistorical dataset of images to develop the recognition system and correlate the trend inconsumer sentiment to operational functioning within our parking garages. Validation will includefield testing of the recognition engine.Responsibilities• Developed and trained an LSTM model leveraging a dataset comprising 417k audio files; Implemented word sequence analysis considering word order using 50-dimensional GloVe pre-trained word embeddings• Enhanced generative accuracy of the sentiment classification model by implementing PyTorch and Tensorflow-based techniques. Achieved an 89% accuracy in emotional classification of AI-generated texts and voices. • Developed and optimized the data transformation and model inference pipeline to enable real-time video sentiment analysis on local FLASH’s kiosks within 5-10s of customer interactions.

[1] Established an innovative open system to thoroughly investigate privacy concerns. This dynamic system was then employed to delve into privacy concerns across diverse scenarios.[2] Pioneered the development of cutting-edge machine learning models capable of accurately predicting fatigability in different clinical populations. Leveraging the profound research on fatigability in multiple sclerosis patients and harnessing the power of transfer learning, our models displayed exceptional performance

• Designed and built a scalable and reliable web app backend infrastructure with databases and server nodes on AWS-EC2 that serves over 1,000 concurrent requests with 0 downtimes. • Implemented a resilient data processing framework using R Shiny and React to perform real-time analysis on user inputs and provide instant feedback and data visualizations with an average latency of 50 ms. • Conducted thorough performance testing and A/B testing to optimize website responsiveness and improve user experience during the migration from local R Studio deployment to AWS, followed by dockerization for integration with the USA NPN platform.

• Developed a robust database system using SQLite and Microsoft Access and collaborated with cross-functional teams to optimize the storage and query efficiency based on data requirements of course registration information for over 80,000 users. • Designed and developed automated data ETL pipelines in Python and reduced data processing latency from 2 days to 10 minutes. • Customized the learning management system (LMS) interface and integrated Qualtrics survey into Canvas, improving user experience and analysis efficiency of the course evaluation infrastructure.

• Leveraged various Google Cloud Platform APIs, including App Engine, Natural Language, and Datastore, to provide guidance for target users with imposter syndrome using Java, JavaScript, HTML, and CSS.• Participated in distributed development to process user input; provided corresponding feedback based on the scores received from users for sentimental analysis.• Contributed to open source software using Github; conducted weekly code reviews; designed newcomponents and web interfaces to improve user experience.

• Tested 5 products manually, observed 12 bugs in total, described defects and differences to expectedresults; individual work resulted in a 33% improvement over the old version.• Conducted competitive product analysis following SWOT framework for 4 different insurance products to help the team locate breakthrough points and product promotion focus; resulted in 1.2x work efficiency.• Identified significant attributes of users; generated reports about test cases, strategy references, and improvement direction for 3 insurance products according to UI diagrams. Assisted the development team in fitting user needs with 100% accuracy.