Curriculum Vitae

Work Experience

May 2025 – Current

Developed a transformer-based cell-phenotyping pipeline for human kidney biopsy images that predicts cell types with 91% accuracy, to investigate novel insights into cellular spatial organization in lupus nephritis and mapping these findings onto murine cell models. Skills: CODEX, Spatial Omics, machine learning.

Jan 2020 – May 2025

Developed and implemented statistical models and Kalman filtering techniques to enhance estimators for real-world dynamical systems on 15 million rows of real-world data from multiple sensors, resulting in an 82% prediction accuracy.
Created a realistic simulator using data-driven probabilistic methods for sparse event simulation along with an empirical data-driven model for preserving real-world characteristics. It further uses mathematical tools like logistic transform and autoregressive functions for smooth, realistic time series evolution.
Integrated an odor simulator with a physics-enabled robotics simulator to develop a framework for designing UAVs for outdoor odor-tracking challenges.

Sep 2017 – Dec 2019

Led a project integrating virtual reality with a real-world robot and applied computer vision techniques and virtual-space to real-world-space mapping, and improved operator learning efficiency by 85%. Skills: Robotics, Computer vision, deep learning, YOLO, virtual reality, Python, C#, C++.

Languages: Python, C++
Tools and Libraries: Pandas, Pyspark, Numpy, Scipy, Scanpy, Cellpose
Software Framework: Pytorch, Pandas, Tensorflow, CUDA
Concepts: Statistical Modeling, Regression Analysis, Bayesian Networks, Deep Learning, Machine Learning, Neural Networks, Data Visualization

Nag Arunava, and Floris Van Breugel. “COSMOS: A Data-Driven Approach to Simulating Odor Encounters for Agents Moving Through Chemical Plumes of Various Scales”, IEEE Open Access (2025)
Nag Arunava, and Floris Van Breugel. “Odour source distance is predictable from a time history of odour statistics for large scale outdoor plumes”, Journal of Royal Society Interface (2024)
Lingenfelter Bryson, Arunava Nag, and Floris van Breugel. “Insect inspired vision-based velocity estimation through spatial pooling of optic flow during linear motion.” Bioinspiration & Biomimetics (2021)

Replaced TIMEGAN (Time-series Generative Adversarial Networks) with GPT models, resulting in an 84% improvement in prediction accuracy for spatial forecasting of time series data from dynamic chemical sensors.

Utilized multi-omics CAR T treatment, and visualized and developed clustering and a convolution neural network-based patient response model based on 18,000 gene expression scRNA features using scanpy and pytorch. Additionally used scGPT cell embeddings to provide 81% accuracy of prediction of patient response and gene–gene interaction.

Made a fast online video summarizer using the Gemini Pro LLM model and the YouTube video API.

A data-driven fast chemical plume simulator was developed to provide a realistic experience for a simulated agent, using gaussian process and Bayesian optimization methods (PyTorch and Python).

Statistical Analysis: Applied Regression Analysis, Advanced Probability (Bayesian Networks)
Machine Learning: Deep Learning, Reinforcement Learning, Application of Graph Theory
Data-Driven Dynamic System Modeling: Control Systems

Invited for participating and speaking at BISCCIT, Bio-Inspired Sensing, Computing, and Control with International Teams, 2023, London.
Presented “Outdoor odor localization” and “Data driven plume simulator” at the AI Dynamics Institute workshop, University of Washington, 2023.
Presented “Mapping outdoor odor plumes using a mobile chemical sensor” at APS 2022, Chicago.