Building the cutting-edge Agentic ML platform that powers semantic search, entity extraction, and summarization.

1. Enhance user search speed using ML/AI and SQL optimization.

2. Build and ship data ingestion pipelines for semantic retrieval.

3. Design and build a data revocation system from end to end.

4. Designa and build on-premise monitoring system.

• Data Processing: Developed video and audio data processing library that is shared by all team members. The library leverages parallel processing that shortens overall data processing time by half.

• Audio Visual Enhancement: Collaborated with a senior research scientist on developing siamese neural network that detects audio-video synchronization error. The model is used to generate meaningful data embedding.

• Implemented time-series data preprocessing and RNN model training API for 100+ battery testing datasets.

• Designed and trained LSTM attention algorithms to predict battery failure time and forecast battery state-of-health with up to 92% benchmark dataset accuracy.

• Investigated data-driven approaches on battery lifetime prognostics and implemented workable demos.

• Implemented web crawlers using Python Scrapy and Selenium package. Scraped 37 sets of battery decay data.

• Created organization Github repository for the lab and maintained 2 projects.

1. Implemented a command prompt tool for F=ma graph database that supports functionalities including parsing database query commands, communicating with database REST server and displaying query results. Successfully deployed the software on client's database.

2. Implemented database automatic I/O testing script. Successfully deployed testing scripts on AWS via Docker containers.

• Developed a data preprocessing pipeline that unifies point cloud data size, orientation, and scale for 10000+ 3D models.

• Developed data denoising method that improved training speed by 10%.

• Acknowledged in “3D Design Using Generative Adversarial Networks and Physics-Based Validation” ASME 2019.

• Collaborated with PHD students on designing a generative adversarial network. Implemented and trained the network by utilizing AWS services and Docker containers.

• Alleviated mode collapse problem by modifying the discriminator structure and introducing Wasserstein loss function, improving average design aerodynamic score by 8%.

• Collaborated in a team of 4 to create web app demos for 3D model showcasing and live 3D model generation.