I am a final year PhD candidate at
Mila,
where I work with
Guillaume Lajoie
and
Yoshua Bengio.
My research spans meta-learning, amortized inference, diffusion models and variational methods,
and large language model optimization — from pretraining to post-training.
During my PhD, I also spent time at various research labs including Google, Meta, Apple, NVIDIA, Morgan Stanley and Uber ATG, where I tackled problems ranging from long-context meta-learning and LLM pre-training
to probabilistic forecasting and distillation.
I am primarily interested in improving the capabilities of current large-scale systems,
from leveraging inductive biases in their architectures to latent variable modeling
(analogously RL) as a framework for improving both reasoning and long-context abilities.
Contact: sarthmit@gmail.com
Education
Mila, Université de MontréalSept '22 - Jan '26 (Expected)Degree: Ph.D. in Computer Science | Fast-tracked from M.Sc.
GPA: 4.3/4.3
Advisors: Guillaume Lajoie, Yoshua Bengio
Indian Institute of Technology (IIT) KanpurJul '15 - May '19Degree: B.S. in Mathematics and Scientific Computing
GPA: 9.7/10.0.
Awarded Suman Gupta Gold Medal
Dean's List for Academic Excellence every year
Experience
Sept '25 - Present
Google
Student Researcher
Hybrid systems for efficient inference in large language models
Sept '24 - Aug '25
Meta
Student Researcher
Iterative Amortized Inference (IAI) for efficient meta-learning
Jul '24 - Sept '24
Apple
Research Intern
Demand forecasting using time-series foundational models for Apple products
Jun '23 - Aug '23
Morgan Stanley
Research Intern
Combining continuous time generative models with autoregressive time series predictors
Jan '23 - Apr '23
York University
Research Intern
Amortizing Bayesian posterior inference using in-context learning
May '22 - Dec '22
NVIDIA
Research Intern
Leveraging synthetic targets for machine translation through knowledge distillation
Sept '19 - Mar '20
Uber ATG
Research Intern
Camera and LiDAR fusion for mapping using autonomous vehicle logs
Scholarships
FRQNT Doctoral Scholarship
UNIQUE PhD Excellence Scholarship
IVADO MSc Excellence Scholarship
UNIQUE MSc Excellence Scholarship
Thinking Machines (Tinker) Research Grant
KVPY Fellowship
Academic Service
Frontiers in Probabilistic Inference: Sampling Meets Learning:
Helped organize the workshop at ICLR'25
