Anil Keshwani
About Me
I am carrying out my M.Sc. thesis on speech language models in André Martins’ group at Instituto Superior Técnico in Lisbon, Portugal with funding from the Center for Responsible AI.
Research Interests
- Multimodality x-modal alignment, speech representation learning, SLU, self-supervised learning
- Translation & Multilinguality multilingual LMs, low resource languages, speech-to-speech translation
- Interpretability explainability, probing, fairness
Experience
Deep Learning Scientist - Translated (2022-24)
- Text-to-speech: Trained & deployed several architectures from scratch in multiple languages
- ML solutions: Built Bayesian A/B testing toolkit for interpretable experiment analysis & a routing model for translation jobs based on document semantics
- Interviews: Ran technical interviews for ML engineering intern candidates
Statistician - FFT Education Datalab (2016-18 & summer 2020)
- Statistical forecasting systems for student grade predictions and value-added analysis
- Causal inference for educational intervention assessment inc. delivery of technical and non-technical reports
- Contributed statistical algorithms to Aspire analytics platform (FFT core product)
Health Intelligence Analyst - Public Health England (2018-19)
- Compiled and published national statistics on dementia and mental health service use
- Automated data pipelines for national dementia & mental health statistics
- Integrated the Mental Health Services Data Set into PHE infrastructure
Education
- M.Sc. Data Science - University Of Rome Sapienza (GPA: 29.9 / 30)
- B.A. Biological Sciences - University Of Oxford (Upper Second Class Honours)
My blog moved here and I publish notes here.