Aman Arya

I work at the intersection of research and engineering, building RAG pipelines, agentic workflows, and MCP-based tooling for machine learning infrastructure.

email · GitHub · Google Scholar · LinkedIn · Twitter ·

Background

2025 – present

Software Development Engineer, Research Foundation for SUNY at Stony Brook. Working with Professor Wei Zhu's group on a full-stack web application for visualizing coastal flood risks, sea level anomalies, and hurricane impacts from real-world observational data.

2023 – 2025

M.S. in Data Science, Stony Brook University. Coursework in machine learning, stochastic models, statistical computing, big data algorithms, and computational finance.

2022 – 2023

Master's thesis student, Indian Institute of Astrophysics, Bangalore. Computed orbits and vertical height distributions for 4006 open clusters in the Galactic disk using Gaia DR3 astrometry, under Professor Annapurni Subramaniam and Dr. Geeta Rangwal. Published in the Journal of Astrophysics and Astronomy.

2018 – 2023

Integrated B.S.–M.S. in Physical Sciences, IISER Mohali. Coursework in statistical mechanics, quantum mechanics, electrodynamics, probability and statistics, computational methods in physics, and mathematical methods. Minor in Data Science, with additional coursework in machine learning and AI.

Publications

Rangwal, G., Arya, A., Subramaniam, A., Singh, K.P., Liu, X. (2024).
"Orbits and vertical height distribution of 4006 open clusters in the Galactic disk using Gaia DR3."
Journal of Astrophysics and Astronomy.
[Google Scholar]

Selected Projects

Hormuz Lens — Next.js full-stack dashboard tracking shipping activity through the Strait of Hormuz, with a LangGraph agent layered on top. Tools wired to pgvector retrieval, SQL queries, and Pearson correlation computations over the underlying dataset.
FinOps Copilot — MCP-based agent for cloud cost analysis across Snowflake, BigQuery, and Splunk. Python and TypeScript, with self-describing tool definitions consumed by an agent loop.
Sea Level Rise Forecast Webtool — Full-stack flood risk visualization platform. Flask, Google Earth Engine, GPT-4, Bootstrap.
Market Regime Detection with HMM — S&P 500 regime detection using Hidden Markov Models, with Forward–Backward, Viterbi, and Baum–Welch implemented from scratch. Airflow data pipeline.
simpleEnsemble — R package implementing Ridge, Lasso, Elastic Net, Random Forest, and Bagging with a unified pipeline.
Stable vs Support Vector Regression — Empirical comparison of NuSVR and CVaR regression, demonstrating mathematical equivalence across 1000 train-test splits.