Mukur Gupta

Mukur Gupta

Be yourself, everyone else is already taken.

— Oscar Wilde

Denoising & Diffusing Intelligence @ Granica

I am a researcher at Granica AI, where we are building foundation models for structured data by challenging the existing notions of discrete and continuous diffusion models. I completed my master's in Computer Science from Columbia University with the Advanced Master's in Research (AMR) advised by Prof. Kathleen McKeown. I completed my Bachelor of Technology from IIT Kharagpur, graduating as the highest GPA holder in my discipline.

As a Graduate Research Assistant in the NLP Lab at Columbia University, I contributed to a DARPA grant to build multi-modal approaches for Cross-Cultural Dialogue assistance in low-resource multilingual videos. Last year, I spent my summer at Apple working on multi-modal complex diagram reasoning models. Before grad school, I worked as an Associate Data Scientist at Gartner, where I designed and engineered solutions for question-answering, high-precision product recommendation, user-interest ranking, and multi-modal information retrieval use cases. I've been recognized and nominated for multiple excellence awards for my work at Gartner.

I am interested in solving challenging use cases on multi-modal video understanding, code generation, adversarial robustness, and diffusion-based generation models. My research has been presented at NAACL, CaLM @ NeurIPS'24, CL4Health @ COLING'24, and IEEE PerCom'24. Most recently, I served as a Technical Program Committee member for the SPT-IoT workshop at IEEE PerCom'26.

Education

  • M.S. in Computer Science (2023 - May 2025)
    Columbia University, New York, NY
    Advanced Master's in Research (AMR) Track advised by Prof. Kathleen McKeown
    Fully funded under MS-GRA (awarded to < 1% students in Engineering School)
  • B.Tech (Hons.) in Mechanical Engg (2017 - 2021)
    Indian Institute of Technology (IIT) Kharagpur
    Graduated as Department Rank 1 (highest GPA holder of stream)
    Bachelor thesis advised by Dr. Pawan Goyal

Research

Sense and Sensitivity: Examining the Influence of Semantic Recall on Long Context Code Reasoning

Adam Štorek, Mukur Gupta, Samira Hajizadeh, Prashast Srivastava, Suman Jana

Under Review, 2025

Proposed SemTrace, a novel technique to measure semantic code recall in LLMs, revealing significant drops in code reasoning accuracy as snippets approach the middle of input context. Identified a disconnect between lexical and semantic recall mechanisms, suggesting current code reasoning benchmarks may underestimate LLM challenges in leveraging in-context information.

placeholder placeholder placeholder

XOXO: Stealthy Cross-Origin Context Poisoning Attacks against AI Coding Assistants

Adam Štorek, Mukur Gupta, Noopur Bhatt, Aditya Gupta, Janie Kim, Prashast Srivastava, Suman Jana

Under Review, 2025

Led to a Security Fix in a Major AI Coding Assistant

Introduced Cross-Origin Context Poisoning (XOXO), a novel attack using semantically equivalent adversarial code modifications to compromise AI coding assistants. Developed GCGS algorithm achieving 83.09% attack success rate across 11 models including GPT-4o and Claude 3.5 Sonnet, demonstrating ineffectiveness of existing defenses.

placeholder placeholder placeholder

AdvSumm: Adversarial Training for Bias Mitigation in Text Summarization

Mukur Gupta, Nikhil Reddy Varimalla, Nicholas Deas, Melanie Subbiah, Kathleen McKeown

NewSumm @ EMNLP, 2025

Introduced AdvSumm, a domain-agnostic adversarial training framework that mitigates bias in text summarization through gradient-guided perturbations at the embedding level. Demonstrated effective reduction of name-nationality and political framing biases without compromising summarization quality, outperforming standard transformers and data augmentation techniques.

placeholder placeholder placeholder

CodeSCM: Causal Analysis for Multi-Modal Code Generation

Mukur Gupta, Noopur Bhatt, Suman Jana

NAACL, 2025 (Oral); CALM @ NeurIPS, 2024 (Poster)

Proposed CodeSCM, a Structural Causal Model for analyzing multi-modal code generation in LLMs using causal interventions and mediation analysis. Discovered that input-output examples significantly influence code generation alongside natural language instructions, quantifying spurious model leanings through latent mediator variables.

placeholder placeholder placeholder

Intent Detection and Entity Extraction from Biomedical Literature

Mukur Gupta, Ankan Mullick, Pawan Goyal

CL4Health @ LREC-COLING, 2024 (Oral)

Conducted comprehensive empirical evaluation showing supervised fine-tuned approaches remain more effective than general-purpose LLMs for biomedical NLP tasks. Demonstrated that PubMedBERT can surpass ChatGPT on NER tasks with only 5 supervised examples, challenging the efficacy of LLMs in domain-specific understanding.

placeholder placeholder placeholder

Avenues in IoT with advances in Artificial Intelligence

Ankan Mullick, Mukur Gupta

SPT-IoT @ IEEE PerCom, 2024 (Oral)

Explored current challenges in Internet of Things and the transformative impact of AI, NLP, and machine learning for domain-specific solutions. Discussed future prospects of deeper fusion between NLP, machine learning, and AI to elevate IoT capabilities and reshape digital interactions.

placeholder placeholder

Curriculum generation using Autoencoder based continuous optimization

Dipankar Sarkar, Mukur Gupta

arXiv Preprint, 2021

Presented Training Sequence Optimization (TSO), a novel curriculum learning approach using autoencoders and continuous optimization to learn optimal training data ordering. Achieved 2 AP empirical gain over random strategy on CIFAR-100 and CIFAR-10 datasets, outperforming existing state-of-the-art curriculum learning algorithms.

placeholder placeholder

Experience

  • Research Scientist — Granica AI, CA
    June 2025 - Present
    Advised by Prof. Andrea Montanari, Stanford University. Developing foundation models for structured data by challenging the existing notions of discrete and continuous diffusion models. Applied Granica's ICLR award-winning data selection research (https://www.granica.ai/research)
    to recommendation systems of a social media giant with 400 million monthly active users, driving significant gains in Click-Through Rate.
  • Graduate Research Assistant — NLP Lab, Columbia University, NY
    December 2023 - May 2025
    Contributed to a DARPA grant. Developed multi-modal large language models (LLMs) for communication changepoint detection in cross-cultural conversation videos. Designed a video-reasoning model by pre-training LLaVa on a next-scene prediction objective, and generated large-scale synthetic video–language data to enhance pre-training. Showed improvements in abductive and counterfactual video reasoning tasks.
  • Applied Scientist Intern — Apple, CA
    June 2024 - August 2024
    Designed a novel diagram question answering module for powering RAG pipeline on complex flowchart images by finetuning multimodal LLMs on synthetic dataset
  • Associate Data Scientist — Gartner, India
    August 2021 - June 2023
    • Patent Under Review: Co-invented peer-based product timeline recommendation and value-tracking system
    • Designed and engineered LLM-based (ChatGPT, Falcon, LLaMA) multi-agent Gartner research question-answering bot with toxicity and vendor bias filters
    • Improved the Pearson correlation of Sentence Transformer by 2% on Gartner data, used in Semantic Textual Similarity and multi-modal search
    • Engineered a multi-stage product recommendation engine using Deep Factorization Machine (Recall-30%) and XGBoost (NDCG@10 - 0.78)
    • Improved NDCG@5 of user-interest ranking by 12% with State-of-the-art Variational Auto-Encoder-based topic modeling on unstructured client inquiry texts
    • Designed a list-wise loss-based learning to rank (LTR) model for user interest ranking using user interaction features (MAP@5 - 0.48)
    • Led rapid prototyping & model monitoring by deploying experimental setup to AWS Sagemaker using MLOps integrations
  • Machine Learning Intern — Hike, India
    May 2020 - August 2020
    Designed a novel autoencoder based Curriculum Learning algorithm with LSTM, led to faster model convergence
  • Data Science Intern — Manthan (Algonomy), IN
    May 2019 - July 2019
    Using machine learning on the clickstream data of mobile app of a global pizza chain to build an intent prediction model to predict the next best time when a customer is most likely to open the mobile app to order a pizza and align all the communications (push notification, SMS, emails) at that time to boost the open rates and reduce the possibility of spamming