Please let me take this opportunity to tell you about myself and my past experiences.
My childhood love for Mathematics and Computers drove me from Tehran, Iran to California, USA where I am passionate about making the world a better place through Natural Language Processing, Machine Learning, and Software Engineering. In my past research and interning experiences, I have had the chance to apply the knowledge that I have gained in my vigorous classes to real life problems and projects.
I am motivated to continuously expand my knowledge and grow in my profession, leveraging my fast-learning abilities, organizational aptitude, leadership qualities, and technical expertise.
I am a recent graduate of EECS at UC Berkeley and currently a Machine Learning Researcher at UC Berkeley's Artificial Intelligence Lab advised by Kurt Keutzer.
I am currently working on Efficient Deep Learning with a focus on Natural Language Processing and Large Language Models. In particular, I am interested in sparsity, quantization, and new training/fine-tuning methods to enable models that can learn more efficiently. I am also developing algorithms to compress large neural network models, focusing on reducing inference time and improving training efficiency.
I have also been exploring AI Agents, focusing on the key components required to build them and the system level decisions that effect their performance (Paper Coming December 2024)
Submitted to MLSys 2025 and on Arxiv.
Accepted as a poster to NeurIPS 2024 and on Arxiv.
Accepted as a poster presentation to ENLSP NeurIPS 2023 Workshop and on Arxiv.
Published to the Journal of Geophysical Research: Planets.
This Repository includes survey of all the Large Language Models, Benchmarks, Normalization techniques, and Activation Functions.
This Repository includes a survey of all key papers in NLP.
Mentor: Professor Kurt Keutzer.
Building efficient LLM-based systems and working on a survey of AI Agents as the first author Collaborated on Squeezed Attention, a technique to accelerate LLM inference in applications where a large portion of the input prompt is fixed. (Submitted to MLSys 2025) Contributed as co-author to KVQuant which enables large context length inference and allows for serving LLaMA-7B with 1M tokens on a single A100! Built an architecture to accelerate generative LLM inference by 40% as co-author for a published paper. Innovated new approaches for efficient deep learning and NLP.
Mentor: Professor Joseph Gonzalez (UC Berkeley)
Completed individual course of study with Prof. Joseph Gonzalez to design project for efficient language models. Fine and prompt tuned language models to build two scientific article-focused chatbots for students and researchers.
Mentors: Prof. Dave Stegman (UCSD) and Dr. Sue Smrekar (NASA, JPL)
Built and analyzed a model of Venus on supercomputers using Python and Fortran with Prof. Dave Stegman. Found that plume-assisted tectonic subduction happens 80% faster than hypothesized while advised by Dr. Sue Smrekar.
Co-authored scientific paper in support of NASA’s Venus VERITAS mission of NASA/JPL.
Designed and implemented python software to solve Nonlinear Differential Equations to speed up analytics by 75%. Simulated home appliance power consumption using the Span Panel data to inform next product iteration. Analyzed the value of electrification technologies with processing usage data of the Panel in Python and Snowflake to predict the best product sales with the goal to generate higher revenue for the company.