Cheng Chen Email & Phone Number

Buildling Liyuan AI, a music Music by integrating suno.ai and LLM. Generating music based on pre-generated voice or user-cloned voice via Singing Voice Conversion & Singing Voice clone and vocal isolation.Features:- Users can specify content and themes to generate music and lyrics, using LLM platforms like GPT-4 to generate lyrics and a custom rhyme checker to improve the quality of Chinese lyrics.- Users can sing along with the music, and the audio will be retained for Singing Voice Cloning.- By applying vocal isolation and Singing Voice Conversion to the music generated by suno.ai, the platform creates music sung by the user's voice.Infra involved:The LLM integration platform used in this project is opensourced at [Github](https://github.com/raceychan/askgpt)AskGPT is a production-ready Dstirbuted, multi-Tenant ML-OPS platform that enables businesses to leverage cutting-edge pre-trained language models like OpenAI's ChatGPT as well as various LLMs model, Support training local models using pre-trained language models such as OpenAI.TechStack: python, c++, FastAPI, postgresql, pulsar, redis, ray, huggingface, SVC, etc.- User and Auth management system, allowing one account for all remote and local model serving, making the model serving layer transparent to user.- Asynchronous IO, parallel computation (500 QPS with 35 model inference response per core). Model serving is done through ray+huggingface/pytorch, with fastapi used.

Shenzhen, Guangdong, China

Shenzhen Lingxing Network Technology Co., Ltd. is SAAS company in Shenzhen. In 2021, the company completed a C round financing of RMB 280 million.As a data engineer, my main responsibilities included Amazon data source acquisition and related microservice architecture design, as well as technical problem-solving for the company's key projects.• Led a team of five backend developers to build a comprehensive data service for Amazon advertising, resulting in a nearly tenfold increase in advertising customer volume, 24K to 200k. • Developed a unified Amazon data collection service, supporting all seller data businesses, with over 10 billion daily requests and collecting more than 2 billion reports (approximately 20TB of data). • built data pipelines with Flink+Debezium, read CDC data and build a cross-service data synchronization service, synchronizing 150 million data points on a daily basis, Utilizing Flink on top of DTS for event tracking, solving the slow query problem for finance-related projects with multi-table joins. The query time was reduced by over 98%. • supervised the team's technical and code quality, built a unit-test system, and conducted code review for key projects as the technical leader

Shenzhen, Guangdong, China

• Building data pipelines, general ETL and data warehousing, using Ray to aggregate data of various sources, including request logs, and business data from DWD layer into ADS and Feature Store. • Feature Engineering using C++, Implementing market-volume measure and calculation algorithm based on related research papers, building a product price optimization system using sub-category data with machine learning algorithms (mainly clustering). • NLP tasks using PyTorch and BERT, corporate with product managers, built a product rating & iterate system based on noun extraction and sentiment analysis on Amazon product reviews. • Analyzing market-specific popular products using PyTorch and Prophet based on full-category data from Amazon sites. • Providing online model serving as a microservice.