- Training infrastructure architect for the AIGC Startup- Diffusion-based model for archtecture rendering- Image Sythesis Evaluation

- I worked on product retrieval for Dynamic Product Advertisement (DPA) to retrieve 20 products for fine-selection from catalogs containing millions of candidates based on user protrait. I studied and optimized for Approximate Nearest Neighbour (ANN) algorithms such as Hierachical Navigable Small World (HNSW) and Fast Ball Tree (FBT).- I designed and implemented a visualization platform for DPA retrieval using Spark, Django and MySQL to parse and analyze large volumns of data. The system went online in a month and has been providing crucial insights for the algorithm team, receiving cross-departmental recognition.- I participated in the formulation, deployment and validation of deep learning models for similar product retrieval, de-biasing and user identification with complicated data pipelines.- I collaborated with members on and outside the team to investigate training data quality. I authored and published a cross-functional reviewed report to pinpoint bottlenecks and guide optimization direction.

- Using pandas, json and re to perform statistical analysis of civil cases regarding purchase contracts and perform feature selection via chi-2 method.  - Using machine learning technique (SVM, LR, RF, NN) with tf-idf & trained word vectors to perform law article prediction on civil judgment data. - Participate in developing a Q&A web-embedded chatbot with Rasa- Using Neo4j and data extracted from civil judgement data to construct a knowledge graph relating law articles and their indicator words.

- Investigate and implement the multi-tasking convolutional neural network proposed in Joint Face Detection and Alignment using Multi-task Cascaded Convolutional Networks.- Deploy the condensed NN on a Raspberry Pi3 and optimize its performance so that it can run real-time with multi-processing.- Apply the network structure to gesture recognition with Pytorch.