Harshit Joshi

josharshit [at] gmail [dot] com

Harshit Joshi
Stanford, CA, US

I’m a first-year CS PhD student at Stanford University with the StanfordNLP group. I completed my Bachelors in IT and Mathematics from Cluster Innovation Centre, University of Delhi in 2021. I’m broadly interested in working on Neuro-symbolic methods, Natural Language Processing and it’s applications in Program Synthesis and Computational Social Science.

Previously, I was a Research Fellow at Microsoft Research with the PROSE group, where I worked on Program Repair and Formula Intelligence in Excel spreadsheets, advised by Dr. José Cambronero and Dr. Vu Le. I have been a Research Software Engineer at Supedio GmbH working on Master Data Management, Medical Entity Resolution and Text Extraction from PDFs. I have interned with the research team at DRDO, Govt. of India, where I worked on an Image Segmentation task in a simulated Gazebo environment. I spent a summer at Chennai Mathematical Institute as a Mathematical Finance Summer Scholar, working on sliding-window approaches for finding abnormal rate of returns. I have also interned with the Data Science team at Cronycle, UK, building a production-ready RSS aggregating pipeline. In the summer of 2018, I was a student developer with Google Summer of Code.

I have also had the good fortune to work with Dr. Sumit Gulwani, Dr. Aditya Kanade, Dr. Rajiv Ratn Shah, Dr. Lucie Flek and Dr. Alicia Nobles.

News (older)

May 10, 2023 I will be joining Stanford University 🌲 for PhD in Computer Science this fall (‘23) !!!!
Feb 7, 2023 Our latest work on LLMs, FLAME: A small language model for spreadsheet formulas, was on the front page of HackerNews and covered by media outlets (TheRegister, techradar)
Nov 20, 2022 Our work, “Repair Is Nearly Generation: Multilingual Program Repair with LLMs” has been accepted at AAAI Conference on Artificial Intelligence 2023 (Core A*)!
Oct 1, 2022 Ran my first 10k! Completed 10.55 Km in 58 mins!!
Sep 8, 2022 Thrilled to share that our work on “Neurosymbolic Repair for Low-Code Formula Languages” has been accepted at OOPSLA 2022, a premier programming languages conference! This is my first work at Microsoft PROSE!

Selected Publications (all)

(*: Equal Contribution)
  1. FLAME: A small language model for spreadsheet formulas
    Joshi, Harshit, Ebenezer, Abishai, Cambronero, José, Gulwani, Sumit, Kanade, Aditya, Le, Vu, Radiček, Ivan, and Verbruggen, Gust
    arXiv preprint arXiv:2301.13779 2023
  2. Repair Is Nearly Generation: Multilingual Program Repair with LLMs
    Joshi, Harshit, Cambronero, José, Gulwani, Sumit, Le, Vu, Radicek, Ivan, and Verbruggen, Gust
    Proceedings of the AAAI Conference on Artificial Intelligence, 37 2023
  3. Neurosymbolic Repair for Low-Code Formula Languages
    Bavishi, Rohan*, Joshi, Harshit*, Sánchez, José Pablo Cambronero, Fariha, Anna, Gulwani, Sumit, Le, Vu, Radicek, Ivan, and Tiwari, Ashish
    In Proceedings of the ACM on Programming Languages (OOPSLA2) 2022

Current and Past Affiliations