Research | Shubhadip Nag

Research Interests

Trustworthy AI LLM Unlearning Explainable AI AI for Education Parameter-Efficient Fine-Tuning Data Subset Selection LLM Fine-Tuning Price Optimization Time Series Forecasting

MS Thesis

IIT Kharagpur · 2026

Controlling AI Trustworthiness Through Data and Prompt Selection

Shubhadip Nag

MS (Research) Thesis · Department of Computer Science & Engineering, IIT Kharagpur · Supervisor: Prof. Dr. Sourangshu Bhattacharya

Read Thesis

Research Publications

NeurIPS 2025

MPSelectTune: Prompt-type Selection for Fine-tuning Improves Concept Unlearning in LLMs

Shubhadip Nag, S. Das, A. Saha, A. Ghosh, S. Das, T. Kumar, S. Bhattacharya, S. Bhattacharya

Reliable ML from Unreliable Data Workshop @ NeurIPS 2025, San Diego, USA

NeurIPS · OpenReview

ICLR 2024

VTruST: Controllable Value Function-Based Subset Selection for Data-Centric Trustworthy AI

S. Das*, Shubhadip Nag*, S. Sharma, S. Bhattacharya, S. Bhattacharya (*Equal Contribution)

DMLR Workshop @ ICLR 2024, Vienna, Austria

Paper · Code

IEEE 2023

A Knowledge-Based Approach for User Profiling from Multilingual Texts in Social Media Platforms

Shubhadip Nag, S. Dalui, S. Ghosh, R. Pal, A. Pal

IEEE IITCEE 2023, Bangalore, India

Paper

Research Projects

Evaluating LLM-Powered AI Tutors

Fine-tuned Qwen2.5-Math models with LoRA and 4-bit quantization to judge the pedagogical quality of AI tutor responses. Ranked 1st in Mistake Identification and First Runner-Up overall at the Datathon @ IndoML 2025.

Concept Unlearning from LLMs

Developing prompt-based strategies to fine-tune LLMs (LLaMA2, LLaMA3, Mistral) using LoRA for concept unlearning. Accepted at the Reliable ML from Unreliable Data Workshop @ NeurIPS 2025.

Data-Centric Trustworthy AI

Proposing VTruST, a controllable framework for training data subset selection balancing fairness, robustness, and accuracy. Accepted at the DMLR Workshop @ ICLR 2024.

Knowledge-Based User Profiling

Sentiment analysis model for multilingual texts covering 10 languages. Published at IEEE IITCEE 2023.

Technical Skills

Languages & MLPython (PyTorch, TensorFlow, Hugging Face Transformers, Scikit-learn, NetworkX, Pandas, NumPy, Matplotlib, Seaborn)

Big Data & CloudApache Spark, PySpark, SQL, Hive, Google BigQuery, GCS, Data Discovery

NLPLLMs, Fine-tuning (LoRA, QLoRA), Prompt Engineering

Graph MLGraph Neural Networks, Complex Network Analysis

ToolsGit/GitHub, Jupyter, LaTeX, CodaLab

Relevant Coursework

Machine Learning, Deep Learning, Natural Language Processing, Artificial Intelligence, Complex Networks, Scalable Data Mining, Probability & Statistics, Linear Algebra.