Hi! 👋
I'm Arpan.
Grad Student at Stanford CS
Building socially intelligent AI systems and improving multimodal LLM systems
Research Poster @ KDD
Stanford Campus
Meta HQ
About Me
I'm a Computer Science MS student at Stanford University and Research Assistant at Stanford AI Lab. I am passionate about building socially intelligent AI systems and improving human-AI interaction.
Previously, I was a Software Engineer (IC4) at Meta working on AI-driven groups and recommendation systems. I completed my BS in Computer Engineering from UIUC.
Research Interests
My research interests lie at the intersection of natural language processing (NLP), human-AI interaction, and graph neural networks (GNNs). I focus on developing socially intelligent agents, detecting corpus-level inconsistencies in LLMs, and enhancing transformers with relational embeddings.
Recent Updates
Experience
Research Assistant
Stanford University • Stanford, CA
- Developing socially intelligent agent systems for non-competitive non-collaborative goals under Prof. Diyi Yang
- Building LLM systems for detecting corpus-level inconsistencies with Prof. Monica Lam
- Enhancing transformers with relational positional embeddings for relational databases with Prof. Jure Leskovec
Software Engineer IC4
Meta Inc. • Menlo Park, CA
- Built new onboarding flows for admins and members leading to increased daily active group creations by 3%
- Streamlined privacy changes and built framework to promote new admins to admin-less groups
- Worked with GenAI in FB Groups to create post summaries and answer agents
Software Engineering Intern
Meta Inc. • New York, NY
- Created infinite scroll comments, polls, and overview tab resulting in stat-sig increase in FB watch time
- Simplified E2E testing framework and documentation for internal languages used by 50+ teams
- Received highest intern rating of 'greatly exceeds expectation (GE)'
Education
Stanford University
Achievements
- •Research Assistant at Stanford AI Lab (SALT, OVAL, SNAP)
- •Focus on AI Systems and Human-AI Interaction
Research
- •Working on large language models and human-AI interaction
- •Developing AI systems for improved user experience
- •Collaborating with SALT, OVAL, and SNAP research groups
Notable Coursework
- •CS 224N: Natural Language Processing
- •CS 224U: Natural Language Understanding
- •CS 234: Reinforcement Learning
- •CS 330: Deep Multi-task and Meta Learning
University of Illinois at Urbana-Champaign
Achievements
- •James Scholar Honors
- •Dean's List (All Semesters)
Notable Coursework
- •ECE 391: Computer Systems Engineering
- •ECE 411: Computer Organization & Design
- •ECE 374: Algorithms & Models of Computation
- •ECE 385: Digital Systems Laboratory
Teaching Experience
- •Course Assistant for ECE 391: Computer Systems Engineering
- •Course Assistant for ECE 313: Probability with Engineering Applications
- •Course Assistant for ECE 210: Analog Signal Processing
- •Honors Lab Instructor for ECE 110H/120H: Intro to Electronics and Computing
Honors & Awards
- •Engineering Open House Best Project Award
- •Knights of St. Patrick Honor
- •Outstanding Academic Achievement Award
Publications
Detection, Categorization, and Comparison of Needs Expressed on Twitter during Crises
Pingjing Yang, Ly Dinh, Hamiz Anjum, Alex Stratton, Arpandeep Khatua, Jana Diesner, Richard Sowers
Submitted at ICWSM 2023
In this study, we use Twitter data to automatically identify who needs what and how types of needs, that we categorized and standardized, have evolved throughout the Ukraine-Russia conflict.
Generating High-Level Article Structure Based on Topic Using Two-stage Seq2seq Model
Arpandeep Khatua, Adit Agarwal, Kevin CC Chang
In Prep for ACL 2023
Multi-stage subtopic generator for long text generation and richer search results.
Hackathon Projects
CourseLoop
DevpostAuto-grading on text extracted from PDF assignments with an OCR pipeline, using NLP in python with 98%+ accuracy in 1 week. Reduced auto-grading time by 50% utilizing better algorithms and libraries.
Kaizen Journal
DevpostBuilt a custom NLP model to classify text based on mental health conditions and a web page for easier access by patients and health-care professionals with an OCR and voice to text functionality.
Mauka - Job Search Portal
DevpostJob search portal by scraping real time information from Google and LinkedIn to help curb increasing unemployment rates due to COVID-19 in developing countries.
Recreational Interests
Oil on canvas - my latest work
President @ Toastmasters International Gavel Club
Looking forward to recreate the macOS wallpaper pictures
Trying out new recipes every weekend!
"If I can't scuba what's this all about?"