Ads Identity Modeling, part of Ads Foundation Modeling team- Lead an ML team to build ads conversion identity models and offsite sequence signals

Built DNN models for ads age targeting, ads audience targeting, ads conversion/user matching prediction, estimated ads conversion, ads conversion embedding (self-supervised learning, Matryoshka representation learning)

IQ R&D team lead and ML engineer (NLP & Deep Learning)- Sentiment analysis (CNN, ELMo, OpenAI GPT, BERT, knowledge distillation with BlendCNN)- Neural aspect extraction (ABAE + SIF, published)- Text summarization (extractive + abstractive, patent pending)- Embedding compression and visualization (PCA + tSNE)- Pretrained embedding serving (in-memory caching, parquet)- Survey question design analysis (XGBoost + feature importance analysis, NLI task, patent pending)- Object detection in video streams (Yolo v3)- Redesigned and scaled data pipelines by 1000x for enterprise clientsFounder of ML Reading Group and ML Design Office Hours to improve ML capabilities organization wide.

Azure Cosmos DB - Development and optimization in query engine (e.g. filter optimization, spatial indexing)- Machine learning for query logical charge prediction- Elastic query for multi-partitioned (> 1000 partitions) database- Low-level C++/SIMD optimizations

Assist students in understanding of data structures and algorithms (ECE 250), and corresponding C++ implementations

Developed multithreaded, non-blocking, and cross-platform Node.js C++ add-on

Linux kernel patching and socket programming (C)Routing protocol development and engineering (C++)