About Me

I am a PhD candidate in Electrical and Computer Engineering at the University of Florida, working in the Machine Learning and Sensing Lab. My research is on self-supervised learning for high-dimensional sensing data, mainly hyperspectral imagery, where labels are scarce and the structure of the data does not look much like natural images.

More recently I have worked on evaluating and routing large language models under real cost and latency constraints, during two Applied Scientist internships at Amazon. Alongside both, I build open-source datasets and tools so other researchers can use this kind of data without absorbing the preprocessing burden first.

I am currently looking for applied scientist and research scientist roles.

What I Work On

Self-Supervised Learning

Masked autoencoders for hyperspectral imagery, where masking whole spectral components beats masking individual bands.

Read more →

LLM Benchmarking & Routing

Measuring correctness, cost, and latency together, then routing each query to the model that best balances all three.

Read more →

Remote Sensing for Ecology

Multimodal NEON datasets and pipelines for tree species classification, released as open-source packages.

Read more →

Hyperspectral Plant-Soil Imaging

Pixel-level soil water content mapping in rhizosphere systems at roughly half-millimeter resolution.

Read more →

Research

Self-Supervised Learning for Hyperspectral Imagery

Doctoral thesis · Machine Learning and Sensing Lab

Hyperspectral images carry hundreds of spectral bands per pixel, and labels are scarce and expensive. Masked autoencoders are the obvious way to learn from the unlabeled majority, but they were designed for natural images, where masking a patch removes information that genuinely has to be inferred.

Spectral bands do not behave that way. Adjacent bands are highly correlated, so masking one is trivially recoverable from its neighbors. Isolated absorption features are the opposite: nothing else predicts them, so masking one makes reconstruction impossible. Band-level masking therefore produces a training signal of wildly uneven difficulty, most of which teaches the model nothing.

My thesis, Beyond Band Masking: Decomposition-Guided Self-Supervised Learning for Hyperspectral Images, argues that the lever is not a cleverer masking rule but the representation space you mask in. Decomposing the spectrum first, then masking whole components rather than individual bands, turns an uneven reconstruction task into a uniform one that exploits the redundancy structure instead of fighting it.

Benchmarking and Routing Large Language Models

Applied Scientist Internship · Amazon, Seattle · Summer 2026

A deployment rarely has a single right model. It holds many, spanning an order of magnitude in price and several-fold in speed, and the model that answers a hard query best is seldom the one that answers an easy query most cheaply. A router turns that spread into an advantage by sending each query to a model chosen for it.

Evaluating routers offline needs an answer key: a record, for every query and model, of what came out and whether it was right. Building one is expensive, so released keys label whichever axes their authors needed, and the axis consistently missing is time. Without it you cannot ask whether a cost-optimal route is also latency-optimal, which is the question an operator actually has.

I built an answer key that carries correctness, dollar cost, and served latency in the same cell, over a pool spanning both open-weight and closed commercial models, and a router that trades the three off at a ratio the operator sets. Measured together, the two resource axes turn out to disagree often enough that treating them as one number leaves real savings unclaimed.

Hyperspectral Imaging for Plant-Soil Systems

Machine Learning and Sensing Lab

Traditional soil moisture sensors give point measurements, but the gradients that drive rhizosphere processes happen at the millimeter scale. I developed the first demonstrated approach for pixel-level soil water content mapping in rhizosphere systems, training a DenseNet-inspired CNN on controlled hyperspectral images of soil with measured water contents, then transferring it to rhizoboxes containing growing peanut and sweet corn.

Two problems had to be solved for that transfer. A systematic intensity difference between setups, which I addressed with a spectral scaling correction, and interference from root spectral signatures, which turns out to be species-dependent: sweet corn's thick roots produce a 73% prediction range reduction when filtered, against 34% for peanut's thinner architecture.

The model reaches R²=0.89 and MAE=1.58% at quadrant level across 28 rhizoboxes, mapping water content at roughly 0.5mm effective pixel spacing, a 26 to 45% improvement over traditional regression methods. The work builds on the HyperPRI dataset, which I led the creation of.

Experience

Industry

  • Applied Scientist Intern — Amazon

    Seattle, WA · June 2026 – August 2026

    Built an answer key measuring correctness, dollar cost, and served latency in the same cell across a pool of open-weight and closed commercial language models, and a query-level router that trades the three off at an operator-set ratio.

  • Applied Scientist Intern — Amazon

    Seattle, WA · June 2025 – August 2025

    Built an agentic system using vision-language models via AWS Bedrock to analyze financial reports, reading plots, tables, and markdown together to generate explanations and recommendations across interconnected business metrics.

  • Google Summer of Code Contributor — NumFOCUS

    June 2025 – August 2025

    Selected from roughly 8% of applicants to build NeonCrops, an open-source pipeline and multimodal dataset of NEON tree crowns for ecological research.

  • Machine Learning Intern — USDA Appalachian Fruit Research

    Kearneysville, WV (remote) · June 2023 – August 2023

    Developed a UNet-based semantic segmentation model separating leafless apple trees from visually similar background trees, supporting orchard automation.

Education

  • PhD, Electrical and Computer Engineering

    University of Florida · 2021 – present

    Thesis: Beyond Band Masking: Decomposition-Guided Self-Supervised Learning for Hyperspectral Images

  • MS, Electrical and Computer Engineering

    University of Florida · 2019 – 2021

  • BS, Electronics and Communication Engineering

    Kurukshetra University · 2014 – 2018

Teaching

  • Supervised Teacher, Fundamentals of Machine Learning

    University of Florida, EEL 5840/4930 · Fall 2021

    Delivered a lecture on PyTorch to 175 students, designed and graded assignments, and held weekly office hours.

  • Instructor, Survey of Machine Learning

    UF CPET Precollege Scholars Program · July 2022

    Designed and taught an introductory course for high-school students, with interactive notebooks covering regression, image processing, and classification.

Prefer the one-page version? Download my resume. For the full academic record, including all publications, presentations, and awards, see my complete CV.

Projects

NeonCrops

An end-to-end pipeline turning NEON airborne imagery into labeled per-crown image arrays across RGB, hyperspectral, and LiDAR. Ships 41,738 tree crowns spanning 234 species and 38 sites, released on Hugging Face.

GitHub →

NeonTreeClassification

Classification pipelines built on the NEON crown data, combining RGB, 369-band hyperspectral, and LiDAR canopy height models for multimodal tree species identification.

GitHub →

HyperPRI

The first temporal RGB and hyperspectral rhizobox imagery dataset for underground plant root study, released publicly alongside UNet baselines for root-soil segmentation.

Dataset →

Repetitive Action Counting

A multi-stream model combining RGB video (MoviNets backbone) with pose estimation (YOLOv7) to count exercise repetitions and classify which exercise is being performed.

Indian Licence Plate Recognition

An ANPR system classifying vehicle, plate type, and characters in a single model at 25 fps, pairing semantic segmentation for localization with LPRNet for character recognition.

arXiv →

F1-tenth Self-Driving Car

Wall following, mapping and localization, and SLAM-based path planning on a one-tenth scale F1 car, running ROS on an NVIDIA Jetson TX-2 with LiDAR and stereo cameras.

F1tenth →

Publications

Journal Articles

  • Graves, S. J., Chowdhry, R., Zhou, M., Harmon, I., Weinstein, B., … & Bohlman, S. A. (2025). Facilitating macrosystem biology with organismal-scale airborne remote sensing: Challenges and opportunities. Functional Ecology.
  • Chang, S. J., Chowdhry, R., Song, Y., … & Zare, A. (2024). HyperPRI: A dataset of hyperspectral images for underground plant root study. Computers and Electronics in Agriculture, 225, 109307. Dataset

Preprints

  • Tanwar, S., Tiwari, A., & Chowdhry, R. (2021). Indian Licence Plate Dataset in the wild. arXiv preprint. arXiv:2111.06054

Presentations

  • Chowdhry, R. & Zare, A. (2022). Hyperspectral Signals in the Soil. Poster, NSF Signals in the Soils Annual Workshop.

Datasets

  • NeonCrops — multimodal NEON tree crown dataset, 234 species and 41,738 crowns across 38 sites. GitHub
  • HyperPRI — hyperspectral images for underground plant root study. Harvard Dataverse