Hi, I’m Sejal Kotian!

I am a Computational Science graduate student at the University of Pennsylvania, focused on AI, machine learning, recommendation systems, and production ML. I enjoy building data-driven systems that move from modeling and experimentation to scalable deployment, monitoring, and measurable product impact.

Sejal Kotian

Education

University of Pennsylvania

M.S.E. in Computational Science & Engineering (Scientific Computing)

Expected Graduation: 2027

  • Focus areas: Machine Learning, Scientific Computing, Optimization, Stochastic Processes
  • Relevant coursework: Machine Learning, Computer Vision, Stochastic Processes, Big Data Analytics, Atomistic Modeling, Advanced Topics in ML
  • Teaching Assistant: Mathematical Foundation for Machine Learning II: Linear Algebra
  • Affiliation: Penn Institute for Computational Science

Indian Institute of Technology, Indore

B.Tech in Metallurgical Engineering and Materials Science

Graduated: 2024

  • Thesis: Graph Neural Networks for Accelerated Materials Discovery
  • Strong foundation in mathematics, materials science, machine learning, and computational modeling

Work Experience

Pickle — AI Engineering & Data Science Intern

Jun 2026 – Present

New York, USA

  • Building recommendation, retrieval, and ranking systems across 600K+ marketplace listings using Python, SQL, multimodal embeddings, and behavioral signals
  • Deploying ML inference on AWS SageMaker and production workflows with AWS Lambda and OpenSearch for embedding updates, vector indexing, and scalable retrieval
  • Designing evaluation, experimentation, monitoring, and production debugging workflows to identify model failure modes and support reliable ML rollouts

Research Assistant – Agentic LLM Text Scoring

Jan 2026 – Present

World Well Being Project, UPenn (Prof. Lyle Ungar) — Philadelphia, USA

  • Designing a scalable agentic AI pipeline with LangGraph and Python, including automated QA validation, logging, and evaluation across 150K+ messages
  • Applied BERTopic clustering and embedding-based modeling to detect user patterns across cohorts and improve scoring results by 11% overall
  • Building reproducible ML workflows with experiment tracking, structured evaluation, and modular validation frameworks for rapid iteration and audits

Deloitte USI — Analyst, Machine Learning & Data Science

Aug 2024 – Aug 2025

Hyderabad, India

  • Built large-scale ML and data pipelines using Python, SQL, PySpark, Hadoop, and AWS across 200K+ SKUs for demand forecasting and predictive optimization
  • Engineered scalable ETL, feature engineering, and predictive modeling workflows across 200M+ records from 20+ enterprise data sources
  • Developed machine learning models for classification and predictive analytics, including a model achieving 92% minority-class recall through tuning, validation, and error analysis
  • Delivered $100M+ in operational impact for U.S. Fortune 500 clients through large-scale analytics, ML modeling, and data-driven decision-making

Aalto University, Finland — Machine Learning Research Assistant

Summer 2024

Espoo, Finland

  • Developed a hybrid data-physics and machine learning framework for hydrogen-tolerant materials, reducing computational runtime by 15%
  • Scaled microstructural simulations across 200+ material structures on CSC high-performance computing clusters using Slurm for parallel execution
  • Built and trained temporal Graph Neural Networks with PyTorch to learn structured representations and predict dynamic stress-strain behavior

INRS, Canada — Mitacs Globalink Research Intern

Summer 2023

Montreal, Canada

  • Generated scientific datasets through DFT and computational modeling for sustainable energy and materials discovery applications
  • Modeled material systems using Graph Neural Networks and PyTorch Geometric for representation learning and scientific property prediction
  • Developed Python and VASP-based graph data augmentation workflows using intermediate structures, improving predictive performance by 5X in limited-data settings

Projects

AI Agent for Sprint Intelligence — YHack'26

Built a multi-step AI agent using RAG, reasoning, tool use, and APIs to convert Slack conversations into structured Jira actions, rank sprint assignees from GitHub expertise signals, and coordinate engineering workflows.

AI Agents RAG FastAPI PostgreSQL Docker

Safety-Aware Multi-Agent RL for Coordination in MiniGrid

Built a safety-aware multi-agent reinforcement learning system using MAPPO and Lagrangian constraints, improving coordination efficiency and task success while reducing safety violations.

Python Reinforcement Learning MAPPO PyTorch

Smart Vision Based Bin Monitoring System

Built a real-time computer vision pipeline for trash detection, tracking, and classification using ROI filtering, ByteTrack, and YOLO pose-based hand removal.

Python PyTorch OpenCV YOLO

CloudPhysician — Vital Extraction using Computer Vision

Developed a fast computer vision and OCR system for automated vital extraction from ECG images, delivering 96% accuracy and sub-second CPU inference for scalable structured extraction.

Python Computer Vision OpenCV OCR

RADAR — Personal Research Recommendation Agent

Built an agentic research assistant that reads recent notes, infers active topics, retrieves relevant papers from arXiv and Semantic Scholar, and ranks personalized recommendations with memory.

Python LLMs LangGraph RAG Retrieval

Graph Neural Networks for Accelerated Materials Discovery

Accelerated crystal structure relaxation using graph neural networks, delivering DFT-comparable energies with 2.51% error for 300+ materials in seconds across 70+ alloy systems.

Python PyTorch GNNs CUDA HPC

About

I am an MSE student in Computational Science at the University of Pennsylvania with a strong foundation in machine learning, data science, scientific computing, and optimization. My work spans recommendation systems, retrieval and ranking, graph neural networks, agentic AI, computer vision, and large-scale production ML systems. I enjoy taking ideas from experimentation and modeling through scalable deployment, monitoring, and measurable product impact.

Contact

Email me at [email protected] or DM me on LinkedIn .