Akshat Kaushal

I am currently pursuing Masters in Computer Science at the University of Pennsylvania with a ‘work to learn’ approach. I have over two years of professional work experience as a software engineer assuming roles of data science intern and full stack engineer. I am proficient in varied technologies related to full-stack development like Java, JavaScript, SQL, JUnit, Selenium, React, Python, Kotlin and in data science I am proficient in Python, R, PyTorch, JAX, scikit-learn, etc.

At Adobe, I developed Agentic-AI Journey Graphs with Graph-of-Thoughts for personalized digital experiences. At Salesforce, I worked on personalization platforms as part of Experience Services under Platform Cloud, contributing to rules as well as AI-based personalization. At OYO Rooms, I worked in the data science team optimizing payment mode for the end user thereby increasing view-to-stay conversion. Recently, I built the first comprehensive open-source JAX implementation of Microsoft’s Aurora weather model.

Email  /  Resume  /  LinkedIn  /  GitHub  /  Google Scholar

profile photo

Research Projects
Microsoft's Aurora in JAX and PyTorch: A Foundation Model for Earth System Forecasting
Open Source Implementation, 2025
Project Page / Code

First comprehensive open-source implementation in JAX with PyTorch interoperability. Translated the 1.3B‑parameter model from PyTorch to JAX, implemented full training loops, and benchmarked LoRA-based fine-tuning strategies.

Brain Tumor Segmentation and Survival Prediction (BraTS 2020)
Research Project, 2021

Ensemble of 3D U‑Net and 2D ResNet architectures for 4D‑MRI voxel segmentation on BraTS20, achieving 32nd place with a 2.7% performance gain over baselines and 15% reduction in training time using AWS S3 optimized pipelines.

Analysis of Effectiveness of Indian Political Campaigns on Twitter
Singhal, K., Sood, K., Kaushal, A., Gehlot, V., Rana, P.S. (2024)
DOI: 10.1007/978-3-031-56700-1_17

Education

University of Pennsylvania

Master of Science in Computer and Information Science (AI Concentration) GPA: 3.88/4.0
Philadelphia, PA, USA | May 2026
  • Coursework:
  • Principles of Deep Learning
  • Machine Learning
  • Computer Vision & Computational Photography
  • Advanced Topics in Deep Learning
  • Networked Systems
  • Big Data Analytics
  • Statistics for Data Science
  • Software Systems

Thapar Institute of Engineering and Technology

Bachelor of Engineering in Computer Engineering GPA: 9.26/10
Patiala, PB, India | Jun 2022
  • Coursework:
  • Advanced Data Structures
  • Machine Learning
  • Database Management Systems
  • Deep Learning
  • Software Engineering
  • Analysis of Algorithms
  • Operating Systems
  • Computer Networks
  • Theory of Computation
  • Computer Architecture
  • Computer Graphics
  • Object Oriented Programming
  • Numerical Analysis
  • Discrete Mathematical Structures
  • Predictive Analytics using Statistics

Work Experiences

Adobe

Software Engineering Intern
San Jose, CA, USA | May 2025 - Aug 2025
  • Created Graph of Thought algorithms for Agentic-AI Journey Graph generation, enabling marketers to translate natural-language into personalized Digital Experience campaigns with integrated guardrails via A2A and MCP protocols, LangGraph and Pydantic AI.
  • Developed intent matching, scoring, and terminology-expansions tailored to Graph-of-Thought models, leveraging FastAPI to inject business context and activate AI workflows through reasoning engine, enhancing Journey Graph scoring and relevance by 34%.
Technologies: LangGraph, PydanticAI, FastAPI, A2A, MCP

University of Pennsylvania

Researcher - Predictive Intelligence Lab
Philadelphia, PA, USA | Jan 2025 - May 2025
  • Single-handedly developed a JAX implementation of Microsoft's 1.3B Aurora Earth model, converting massive PyTorch weights and code into a GPU-optimized setup with mixed-precision training, checkpointing, and kernel fusion for efficient model training.
  • Engineered parameter-efficient fine-tuning pipelines with LoRA integration and multi-step autoregressive training, managing terabytes of ECMWF data through hybrid PyTorch-JAX dataloaders while reducing memory footprint by 40%.
Technologies: JAX, PyTorch, LoRA, XLA, ECMWF

Salesforce

Member of Technical Staff
Hyderabad, TG, IN | Jul 2022 - Jul 2024
  • Next-Gen Personalization Platform: Architected and developed type system databases, CRUD operations, validation layers, and data mapping with Data Cloud, processing 200K+ events per second and achieving 40%+ platform adoption rate for enhanced personalized customer experiences.
  • Product Analytics & Metrics: Researched and implemented comprehensive adoption tracking mechanisms for Experience Cloud products, integrating unsupported type systems through validation layer instrumentation and cross-referencing with Salesforce Unified Data Dictionary.
  • Performance Optimization: Designed and implemented performance testing suite using EKG and Armada frameworks, optimized client-side caching strategies, and improved code deployment efficiency, resulting in 24% overall product performance improvement.
  • Production Reliability: Maintained 90%+ code coverage through comprehensive unit and functional testing, ensuring 99.5% system reliability. Led troubleshooting for 30+ Sev1 and 10+ Sev0 incidents within SLA requirements, supporting 45K+ enterprise customers through 14 on-call rotations.
Technologies: Java, JavaScript, Spring Framework, React, Armada, JUnit, Selenium, Splunk, Prometheus, Kubernetes

OYO Rooms

Data Science Intern
Gurugram, HR, IN | Jan 2022 - Jun 2022
  • MLOps Pipeline Development: Designed and productionized end-to-end ML pipelines using PySpark, AWS, and CatBoost for hotel booking prediction and revenue realization, processing 6M+ data points with calibrated classifiers, improving view-to-stay conversion rate by 8%.
  • Data Engineering & Analytics: Architected complex data extraction workflows using SQL, Hive, and PyHive from OYO's Metabase infrastructure, integrating multi-dimensional hotel and user features to create robust feature sets for predictive modeling.
  • Model Optimization: Implemented hyperparameter tuning with advanced feature engineering (hotel location density, customer booking patterns) and post-model calibration using Platt scaling, optimizing personalized payment recommendation systems.
Technologies: Python, PySpark, AWS, Scikit-learn, CatBoost, MLOps, PyHive, SQL, Metabase

Thapar University

Research Intern
Patiala, PB, IN | Mar 2021 - Aug 2021
  • Medical Image Analysis: Developed brain tumor segmentation and survival prediction framework for 4D-MRI analysis using BraTS20 dataset on NVIDIA DGX systems, achieving 32nd place globally with 2.7% performance improvement over baseline architectures.
  • Deep Learning Architecture: Designed novel ensemble model combining 3D U-Net fully convolutional networks with pre-trained 2D ResNet architectures, optimizing feature extraction pipelines and reducing training time by 15% through efficient AWS S3 data handling.
Technologies: Python, TensorFlow, Keras, AWS S3, NVIDIA DGX, Medical Imaging, 3D CNNs

Personal Projects

Distributed Deep Learning Network for 3‑Dimensional Mapping

Course Project
Philadelphia, PA, USA | Oct 2024 - Dec 2024
  • Introduced neural network training inspired by the DiNNO framework, developing CADMM-based optimization and weight-averaging consensus to balance local vs global model learning.
  • Evaluated on MNIST and 2D mapping, extended to 3D; addressed scalability, convergence, communication efficiency, and distributed training complexity.

Vision‑Based Auto‑Zooming Cameraman

GitHub Project
Philadelphia, PA, USA | Sep 2024 - Dec 2024
  • Engineered an autonomous camera system that adjusts zoom and focus in real-time for sports using YOLOv11 for detection, Gaussian blur heatmaps, and temporal smoothing.
  • Built post-game analytics: player clustering, team heatmaps, ball possession, and action timeline via an interactive dashboard.

This template was borrowed from Jon Barron Last update: Aug.2025