Applied AI shaped by research and engineering.

Currently building AI agents for data science tasks, with earlier research in generative speech, foundation models, and natural spoken interaction.

• Developing agentic systems that reason through multi-step data science workflows and work with analytical tools.
• Led MagpieTTS and MagpieTTS-LF research for robust, expressive, and coherent long-form speech generation.
• Developed full-duplex speech-to-speech modeling for conversations with continuous listening and interruption.
• Worked on neural audio codecs, transducer-based TTS, lightweight speaker adapters, and unified speech-language models.

Built and optimized deep-learning systems across computer vision and graph neural networks.

• Explored heterogeneous graph transformers and distributed GNN training with DGL and PyTorch.
• Optimized computer-vision architectures including Mask R-CNN and EfficientNet.
• Developed production-oriented distributed training and model optimization workflows.

Developed multimodal anomaly-detection and product-description quality systems for Google Shopping, combining text, image, and categorical signals.

Researched OCR correction for historical news using language models and neural sequence modeling, supporting analysis of responsible terrorism coverage.

Developed a statistical NLP model for computational stylometry and authorship patterns in scientific writing.

Worked on enterprise software and the migration of large-scale transaction systems to Microsoft Azure, building a foundation in systems architecture and production engineering.