Arman Pouyaei Start a conversation

I’m a scientist with industry experience. I build production-level tools — hazard models, AI emulators, GPU ensembles, petabyte pipelines — that turn weather data into decisions.

Selected work.

Climate hazard modeling

Physics-based and machine-learning models for climate hazard perils — severe convective storms (hail, thunderstorm wind, tornado), subsidence and vertical land motion, wildfire, flood, and more. Most recently: leading MSCI’s blizzard, dust-storm, and sandstorm modeling program end to end — architecture, AI weather emulation, GPU ensembles, validation, and production delivery.

MSCI · New York · now, PyTorch · JAX · CUDA · extreme-value theory

The cone of uncertainty, live

Every forecast is an ensemble. Drag the lead time and watch the tracks fan out, uncertainty grows the further ahead you look.

now
72 h
F5 tornado, Elie, Manitoba, 2007
Tornado
Hailstones, Brisbane hailstorm, 2014
Hail
Lick Fire, Umatilla National Forest, at night
Wildfire
Flooded houses in Chertsey, water up to the garden fences
Flood
Approaching haboob dust storm, Cimarron County, Oklahoma
Dust storm
Blizzard conditions
Blizzard
Six of the perils, modeled — tornado, hail, wildfire, flood, dust storm, blizzard.

Petabyte data pipelines

Architected containerized ETL on AWS Batch and distributed compute with Dask and Slurm on HPC, serving Zarr, Parquet, and NetCDF to downstream analytics and client-facing products.

Platform · AWS Batch · S3 · Docker · Dask · Slurm on HPC

Partition pruning

One dataset, 48 partitions — 12 months × 4 regions, 25 TB each. Slide from a full table scan to a pinpoint query: each stop adds a filter, and the engine skips every partition it doesn’t need. Reading 1 partition instead of 48 is what makes petabyte-scale affordable.

SELECT * FROM storms

Rows of server racks in a data center
Where this kind of pipeline lives. BalticServers data center.

Open science, inspectable

Built and open-sourced multiple projects, such as a Lagrangian trajectory framework, adopted by external research groups for scientific purposes.

Open science, Geospatial · Fortran · atmospheric science

The ensemble, live

Ten ensemble trajectories in gray, the median in blue — one parcel, ten plausible paths. Drag time forward to follow the plume downstream, or backward to trace where the air came from. That forward/backward idea is the heart of the Lagrangian particle tracking tool.

t₀
t = 0 h
A 1650 wind-rose chart mapping the winds, by Jan Janssonius
Tracing winds, then and now. Wind-rose chart, Jan Janssonius, 1650.
The Kraken supercomputer at Oak Ridge National Laboratory
Heavy iron for heavy ensembles. Kraken supercomputer, Oak Ridge National Laboratory.

From research to production.

Based

Princeton, NJ and New York, NY

Doctorate

Atmospheric Science, University of Houston, 2022

Now

Senior Data Engineer (Data Science), MSCI

I’m a scientist working in industry: a PhD in Atmospheric Science turned production engineer, hardening research-grade methods — data assimilation, ensembles, extreme-value statistics — into tools that run at planetary scale.

Experience

2026, now

Senior Data Engineer (Data Science)

MSCI · New York

Lead global severe-storm hazard modeling; build production ML, ensemble forecasting, extreme-value analysis, and AI-assisted diagnostic workflows across hundreds of millions of records.

2025, 2026

Data Engineer / Research Scientist

First Street · New York

Designed global ML workflows, automated QA and anomaly detection, and petabyte-scale AWS pipelines for climate-risk analytics before First Street's acquisition by MSCI.

2023, 2025

Postdoctoral Research Associate

NOAA GFDL / Princeton University

Led a modeling workstream across four research groups and 8+ scientists; benchmarked predictive components, documented uncertainty, mentored researchers, and published high-dimensional climate analysis.

2018, 2023

Graduate Researcher and Postdoctoral Fellow

University of Houston

Developed Bayesian data assimilation, particle transport, data-fusion, and deep-learning methods; earned a PhD in Atmospheric Science and open-sourced research tooling adopted by the wider community.

Hurricane Iota, November 2020, GOES satellite loop. CSU/CIRA and NOAA.

Technical signal.

30+ publications

Seven first-author papers across climate dynamics, atmospheric chemistry, data assimilation, and modeling systems.

Invited speaker

NASA GISS Seminar Series, NOAA GFDL Seminar Series, AGU, AMS, and CMAS.

Technical leadership

Coordinates engineers, analysts, scientists, and product partners around measurable requirements and delivery milestones.

Community service

Lead convener for an AGU 2026 session and peer reviewer for nine scientific journals.

Building something ambitious with data? Write about modeling, AI platforms, or decisions at scale.

arman.pouyaei@gmail.com
Princeton, NJ · New York, NY