Pratyush Muthukumar
pratyush.muthukumar@gmail.com
EXPERIENCE
NVIDIA
| Generative AI Data Scientist
October 2023 – Present | Santa Clara, CA
•
Developed GPU-parallelized
NVIDIA DALI
data pipeline to preprocess long context
video pre-training data with
7X
increase in throughput
•
Implemented training & inference scripts for video multimodal vision-language model
(VLM) leveraging
Megatron-LM
and
TensorRT-LLM
and trained to convergence
exhibiting
4X
speedup across H100 & A100 GPU hardware and
48%
decrease in error
vs image-only VLM in NeMo
•
Developed
multi-turn synthetic question answer data
generation algorithms to
curate training data for training RAG models in the NeMo LLM service at scale
•
Led design of CLI-based evaluation tool to monitor & visualize embedding space drift
with
t-SNE
&
PCA
and uniformly subsample by embeddings via
Voronoi tesselations
TESLA
| Data Engineer Intern
June 2023 – September 2023 | Palo Alto, CA
•
Developed dashboard tools to execute
Spark / Trino
queries and visualize vehicle fleet
data with
32 TB daily ingestion rate
hosted on AWS
•
Produced time-series reliability predictive model to query vehicle metrics via
SQL
for
harmful trend detection in vehicle firmware updates and configured ETL data
pipelines via
Airflow
.
LAWRENCE BERKELEY NATIONAL LABORATORY
| Research Intern
February 2021 – June 2023 | Berkeley, CA
•
Developed robust monitoring tools through the SDNs for End-to-End Networked
Science at the Exascale (SENSE) project
•
Deployed AI-powered network monitoring system on the SLAC and NERSC labs using
Prometheus
and
Grafana
frameworks to collect and query metrics on end-to-end
dataflows of over
2 petabytes
.
NASA JET PROPULSION LABORATORY
| Research Associate
August 2019 – December 2022 | Los Angeles, CA
•
Used a sequential encoder-decoder
ConvLSTM
model to predict atmospheric
remote-sensing imagery of PM2.5 spatially continuously over LA with results showing
over 30% reduction in error
and
24x temporal resolution
over existing approaches.
•
Applied Generative Adversarial Network to predict aerospace stock trends through
financial text and numerical data showing
over 65% decrease in error
over first-day
predictions and
over 70% decrease in error
over 30-day predictions.
PROJECTS
EDVENTURE
| Stanford + Meta Corporate Project
Jan 2023 – June 2023 | Stanford, CA
•
Designed VR web app companion utilizing
React
,
HTML/CSS/JS
, and
PlayCanvas
to
enable virtual exploration of historical sites with the Meta GHA platform
•
Prototyped VR knowledge sharing tools with feedback from the Meta
New Product
Experimentation
Team and presented at Meta HQ
SATELLITE IMAGE WELL PAD MAPPING
| Stanford AI Laboratory (SAIL)
Jan 2023 – June 2023 | Stanford, CA
•
Developed multi-task object detection model for mapping methane-emmision sources
from global imagery through the Stanford ML Group advised by Professor
Andrew Ng.
•
Achieved all-time high mAP scores in
8 of 10
test regions while preserving scalability
EDUCATION
STANFORD UNIVERSITY
M.S. in Computer Science
Specialization in AI
Graduated June 2024
Supported by NSF GRFP
GPA: 3.9 / 4.0
UNIVERSITY
OF
CALI-
FORNIA, IRVINE
B.S. in Computer Science
with minor in statistics
Graduated June 2022
GPA: 3.9 / 4.0
SKILLS
PROGRAMMING
3+ years:
Python • SQL • HTML/CSS
2+ years:
Java • C++ • MATLAB
TECHNOLOGY
AI • AWS • Linux • Git
Spark • Web Dev • Tensorflow
Docker • Trino • Networking
Kubernetes • Keras • PyTorch
COURSEWORK
UNDERGRADUATE
Artificial Intelligence
(Teaching Assistant)
Data Structures
Object-Oriented Programming
Calculus I-III
Differential Equations
Linear Algebra
Probability Theory
GRADUATE
NLP with Deep Learning
Principles of Robot Autonomy
Experimental Robotics
Machine Programming
Financial Trading Strategies
LINKS
Github://
PannuMuthu
LinkedIn://
pratyush-muthukumar
Portfolio://
pratyushmuthukumar.com