Hi! My name is Julian. I live in San Francisco and work as a Senior Research Scientist in the Foundational Research Unit at Google DeepMind (GDM), which I joined in 2024 after over a year in the former Perception team at Google Research. Before that, I was a Senior Applied Scientist at Amazon AWS AI, where I worked for over five years.
At GDM, I work on AI models for generative audio, with a focus on speech and dialogue. This includes:
- Live, audio-native inputs/outputs in Google Gemini
- Interactive multimodal assistants via Project Astra
- Creative music co-creation via Lyria RealTime
- Research on spoken language models and frontier problems in audio synthesis.
More broadly, my expertise is in deep learning for human language, i.e., natural language processing (NLP). My publication topics include:
- Speech generation and synthesis: SpeechSSM · Spectron · Very Attentive Tacotron · Mono-to-Binaural
- Large pretrained language models: Masked LM Scoring · Transformers without Tears · Meta-Learning the Difference
- Speech recognition and self-supervision: Transformer + CTC · Align-Refine · DeCoAR · BERTphone · Attention Failures in ASR
- NLP for low- to zero-resource languages: Unsupervised Translation · Prompting with Phonemes · 0-shot E2E SLU · 0-shot X-lingual Evals,
though I’m always interested in pure mathematics, as a subject and as a benchmark. My Erdős number is 2.
At AWS, I contributed to AI cloud services via both production-scale and custom language models, building automatic speech recognition (ASR) systems, and fast acoustic architectures. I graduated from Harvard with an honors degree in Mathematics and a secondary in Computer Science, during which I variously held summer research fellowships (U.Chicago; Cambridge), took a leave and worked as a full-time mobile developer (Top Hat), was a teaching fellow (CS50) and course assistant (Differential Topology; Galois Theory), held paid internships (J.P.Morgan S&T; HubSpot), and directed a non-audition show choir. I hail from the Philippines and from Canada.
Contact me@my-domain-name-goes.here!