Vitthal Bhandari

I am a Master's student in Computational Linguistics at the University of Washington UW. Before my Master's I spent more than 4 years in the banking industry as a generalist software engineer working at American Express, Standard Chartered Bank, and PayPal. I completed my Bachelor's in Computer Science and Engineering from BITS Pilani where I also did a minor in Data Science and worked with Prof. Poonam Goyal and Prof. Sundaresan Raman.

Email  /  CV  /  Scholar  /  Twitter  /  Github  /  LinkedIn  /  Blog

profile photo

Research [ minimize ]

I'm currently interested in the responsible use of AI spanning across AI4Science, AI for Social Good, and AI Safety.

Here is some more information about me.

  • What kind of research am I interested in? I am an NLP researcher who applies ML to solve sociotechnical issues using human-centered approaches. Few of my major previous projects are:
    1. Leveraging Pretrained Language Models for Detecting Homophobia and Transphobia in Social Media Comments - Link - ACL 2022 workshop paper
    2. On the Challenges of Building Hate Speech Datasets - Link - preprint
    3. LGBQTweet - A community-sourced dataset for detecting hate against sexual and gender identity minorities
    4. Studying the evolution of attitudes toward gender identity minorities over time across partisan leanings in popular political podcasts - Code
  • What am I currently working on?
    1. Evaluating the cultural competence of LLMs - Project for LING 575 (Societal Impacts of Language Technology)
    2. Stanford CS 336: Language Modeling from Scratch
  • Ideas I am interested in exploring: I do have a few broad overarching themes in my mind (non-exhaustive list):
    1. AI4Science: how can we build trustworthy large language agents, capable of generating novel "compelling" hypotheses?
    2. AI4SG: How do LLMs encode cultural information? Is their "cultural sensitivity" scalable across geographies? Are we proactively identifying stakeholders who are at imminent harm from deployment of these language technologies?
    3. AI Safety: I am interested in assessing the harms of emergent misalignment from a human-centered lens
  • Why am I a great hire? I am a generalist software engineer with 4+ years of experience across American Express, PayPal, and Standard Chartered. Here are some notable highlights and projects I have under my belt:
    1. At Amex, I created a language translation feature (HuggingFace, Sanic, Flask) using the Marian MT framework for translating chats between English and Spanish
    2. At Standard Chartered I led a team of 4 in developing an API management framework (TypeScript, React, Flask) and creating configurable APIs (1.3K+ APIs) to fetch, add, and update data from Oracle SQL and Dremio data lakes
prl On the Challenges of Building Datasets for Hate Speech Detection
Vitthal Bhandari
Preprint

This paper presents a comprehensive framework that standardizes the dataset creation pipeline across seven critical checkpoints by identifying systemic challenges in hate speech dataset creation.

arXiv
blind-date Leveraging Pretrained Language Models for Detecting Homophobia and Transphobia in Social Media Comments
Vitthal Bhandari and Poonam Goyal
ACL 2022 Workshop on Language Technology for Equality, Diversity and Inclusion

I contributed to a shared task focused on identifying homophobic and transphobic content in YouTube comments by implementing basic classifiers using multilingual pre-trained language models to analyze English, Tamil, and code-mixed datasets.

Paper | Code
clean-usnob Reviewing the collaborative role of Image processing in retinal imaging
Rehana Khan, Vitthal Bhandari, Sundaresan Raman, Abhishek Vyas, Akshay Raman, Maitreyee Roy and Rajiv Raman
Teleophthalmology and Digital Health: A Practical Guide to Applications, Springer Nature

Paper

Coursework
LING 570: Shallow Processing Techniques for Natural Language Processing
LING 575: Societal Impacts of Language Technology
Stanford CS 336: Language Modeling from Scratch
Harvard CS 2881: AI Safety
Stanford CS 234: Reinforcement Learning

Credits of this template go to source code.