Domenic Donato
About
Detail
CEO / Co-Founder
Orlando, Florida, United States
Résumé
Jobs
verified_user
0% verified
CEO / Co-FounderAttuned IntelligenceMay 2024 - Current (2 years 4 months)A platform purpose-built for hospital and health system call centers, answering every patient call instantly and safely resolving up to 70% of all interactions.- VP of TechnologyAssemblyaiOct 2022 - May 2024 (1 year 8 months)Helped AssemblyAI grow from Series A through Series C, a 5x in ARR. Scaled the technology organization from 15 to over 50 FTEs. This encompassed the research, engineering, program management, security, and IT teams. Contributed by leading the technology teams, producing flagship AI models, establishing partnerships with AWS and GCP, and collaboratively strategizing on product roadmap, pricing/packaging, and revenue forecasting.
- Research LeadAssemblyaiMar 2022 - Oct 2022 (8 months)Set research direction and best practices for the research teams—Speech Recognition and Language Modeling.
- Research EngineerGoogle DeepmindAug 2017 - Mar 2022 (4 years 8 months)Research Scientist, Software Engineer, and Project Lead. Specialize in deep learning research applied to natural language sequence-to-sequence modeling and have multiple publications at top tier venues in this area. Have successfully led multiple teams of various sizes on both research and software engineering projects. Able to identify pursuits most likely to lead to concrete results and make sure we "fail fast" when exploring new ideas.
- Software EngineerGoogleOct 2014 - Aug 2017 (2 years 11 months)Contributed to core Google email, sms, and push notification infrastructure; a high throughput system with strong SLA. Developed features that enabled customized email marketing workflows for Ads and worked with many other Google teams to offer customized notifications that were easy to opt out of.
- ConsultantSlalomFeb 2014 - Oct 2014 (9 months)Helped clients transition from desktop software solutions to cloud based solutions.
- Research EngineerChapman UniversityJan 2012 - Feb 2014 (2 years 2 months)Responsible for complete lifecycle of multiuser applications created to support the research of 7 university faculty members. Applications are used to conduct user driven simulations of an economic environment such as an asset market (eg. stock market).
- Graduate Research AssistantChapman UniversityAug 2010 - Jan 2012 (1 year 6 months)Created/tested software for economic experiments as well as wrote instructions for some of the experiments. Carried out an extensive research/consulting project, which required the training of 200 subjects and over 60 experiment sessions. In addition, my colleague and I were responsible for analyzing the data and producing the report that was submitted to the organization that requested the research. Conducted well over 100 software based economic experiments as a graduate research assistant.
- Network AdministratorGlobal Capital MarketsAug 2009 - Aug 2010 (1 year 1 month)Maintained the internal network, backups, file repository, and performed basic IT troubleshooting for Principles in Irvine office. Assisted with presentation packages sent to potential suitors for M&A projects. Responsible for creating list of suitors that may show interest in client being represented by Global Capital.
Education
verified_user
0% verified
- M.S, Economic System DesignChapman UniversityJan 2010 - Dec 2012 (3 years)
- B.S, Business Administration; Finance and MarketingChapman UniversityJan 2005 - Dec 2009 (5 years)
Publications
verified_user
0% verified
- IEnabling Arbitrary Translation Objectives with Adaptive Tree SearchICLRFeb 2022Coauthored publication on using a variant of MCTS for sequence to sequence decoding that enables using final sequence level metrics to guide search in autoregressive models. We tried many different methods for scoring the quality of a final translation when a reference translation is not available (eg. production inference). The main issue encountered was that cross entropy trained transformer models are poorly calibrated, which results in worse translations as the search budget increases even though the model score (search objective) improves. My key contribution was to fine-tune the models using minimum risk training (MRT). This exposed the model to its own output during training, finally enabling the MCTS algorithm to outperform Beam Sea
- DScaling Language Models: Methods, Analysis & Insights from Training GopherDeepMind WhitepaperJan 2022Member of DeepMind large group effort (60+ members) to scale and better understand large pre-trained language models. I focused on using few-shot learning to indicate the probability that a statement was supported by the supplied evidence. Through prompt engineering and constrained response output, we were able to achieve a new state of the art on the FEVER fact checking dataset.
- Diverse Pretrained Context Encodings Improve Document TranslationACLAug 2021Augmented transformer architecture with additional self-attention modules to incorporate document context (previous and next sentences) from BERT and PEGASUS pre-trained language models. This resulted in more cohesive document level translations.
- EThe DeepMind Chinese–English Document Translation System at WMT2020EMNLPNov 2020This paper describes the DeepMind submission to the Chinese→English constrained data track of the WMT2020 Shared Task on News Translation. Our final system provides a 9.9 BLEU points improvement over a baseline Transformer on our test set (newstest 2019).