Mohammad Saqib Hasan

Ph.D. Candidate in Computer Science

Hi everyone! I am Saqib, a fifth year Ph.D candidate at department of Computer Science at Stony Brook University, working at the Language Understanding and Reasoning (LUNR) lab under Professor Niranjan Balasubramanian. My research focuses on improving the code generation ability of language models using a plethora of techniques ranging from synthetic teacher distillation to preference optimization. Outside of work, I enjoy exploring the latest advancements in current affairs and the restaurant scene around my current town.

Work Experience

Applied Scientist Intern

Amazon

Developed GenAI systems for automated SOP creation and analysis of large-scale databases using advanced retrieval augmented generation techniques, problem-centric tool creation and multi-agent framework.

RAG Multi-Agent GenAI

Research Engineer

Bangladesh University of Engineering & Technology (BUET)

Conducted research on developing novel weight initialization for deep learning models, improving the paraemeter efficiency of neural networks using compression techniques and integrating active learning optimization strategies for efficient fake news detection.

Neural Compression Weight Initialization Active Learning

Education

Aug 2021 - Present

Ph.D. in Computer Science

Stony Brook University

Advisor : Niranjan Balasubramanian

Dissertation : Advancing Code Generation through Context

Jul 2014 - Oct 2018

Bachelor of Science in Computer Science

Bangladesh University of Engineering and Technology (BUET)

Graduated with Honors (Rank 20 out of 138)

Advisor : Muhammad Abdullah Adnan

Thesis : Analysis of Weight Distribution and Initialization in Neural Network Inspired by Neuro-science

Publications

Teaching an Old LLM Secure Coding: Localized Preference Optimization with Distilled Preferences 🔗

Mohammad Saqib Hasan, Saikat Chakraborty, Santu Karmaker, Niranjan Balasubramanian

The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

Handling Open-Vocabulary Constructs in Formalizing Specifications: Retrieval Augmented Parsing 🔗

Mohammad Saqib Hasan, Sayontan Ghosh, Dhruv Verma, Geoff Kuenning, Erez Zadok, Scott Smolka, Niranjan Balasubramanian

First Conference on Language Modelling (COLM 2024)

Truth or lie: Pre-emptive detection of fake news in different languages through entropy-based active learning and multi-model neural ensemble 🔗

Mohammad Saqib Hasan, Rukshar Alam, Muhammad Abdullah Adnan

2020 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM 2020)

Compressed neural architecture utilizing dimensionality reduction and quantization 🔗

Mohammad Saqib Hasan, Rukshar Alam, Muhammad Abdullah Adnan

Applied Intelligence, Springer

Neuro-Scientific Analysis of Weights in Neural Networks 🔗

Mohammad Saqib Hasan, Rukshar Alam, Muhammad Abdullah Adnan

International Journal of Pattern Recognition and Artificial Intelligence (IJPRAI), World Scientific

MuSciClaims: Multimodal Scientific Claim Verification 🔗

Yash Kumar Lal, Manikanta Bandham, Mohammad Saqib Hasan, Apoorva Kashi, Mahnaz Koupaee, Niranjan Balasubramanian

ArXived, In Review

Projects

RAG Agents for Scientific Claim Verification

Built a real-time streamlit application that uses LLM agents and retrieval augmented generation to verify scientific claim validity. The system retrieves relevant scientific documents and generates responses based on the retrieved context.

LLM Agents RAG Streamlit

Ability of LLMs when Pretrained on Sequences Other than Natural Language

Analyzed the ability of LLMs pretrained on sequences other than natural language. Observed that LLMs pretrained on structured sequences such as code and scientific reports exhibit stronger performance on downstream tasks compared to those pretrained on natural language.

LLMs Fine Tuning

Weakly Supervised Active Learning for Named Entity Recognition

Developed a novel training methodology that combines weak supervision and active learning to enhance named entity recognition performance with minimal annotated data.

Active Learning Weak Supervision Named Entity Recognition

Understanding Covid-19 Vaccination Data Across US Counties

Analysed the various elements affecting the Covid-19 vaccination rates across different counties in the United States. This project involved analyzing vaccination rates using socio-economic variables & tweet sentiment, and developing models to predict future vaccination rates for each county.

Data Science Sentiment Analysis Time Series Prediction

Curriculum Vitae

Download my complete curriculum vitae for detailed information about my academic background, research experience, and accomplishments.

Download CV (PDF)

Contact Me

Let's Connect

Always open to discussing research, collaborations, and new opportunities in NLP.

CS Department • Stony Brook University • Stony Brook, NY