Md Mahadi Hasan Nahid
I am a graduate student in Computing Science at the University of Alberta, Canada, under the supervision of Professor Davood Rafiei. Currently, I serve as a research assistant at the AI and Database Lab, where I actively contribute to cutting-edge research initiatives.
My research interests span across Natural Language Processing, Information Retrieval, and Large Language Models. I am passionate about exploring innovative approaches to enhance language understanding and retrieval systems.
Prior to my graduate studies, I completed my Bachelor’s in Computer Science and Engineering from Shahjalal University of Science and Technology (SUST), Bangladesh.
🧭 🎓 🤖 Research Overview
My research explores how large language models (LLMs) can understand and reason over structured and semi-structured data such as tables and databases—domains traditionally dominated by symbolic systems. I aim to build frameworks that integrate understanding, retrieval, and reasoning with the adaptability and generalization power of large language models. The unifying vision behind my work is to develop trustworthy, interpretable, and agentic AI systems capable of performing complex reasoning, grounding, and decision-making across structured data and natural language.
The journey began with two foundational studies—TabSQLify and NormTab—that examined the internal bottlenecks of LLMs in tabular reasoning tasks.
TabSQLify introduced a table decomposition strategy that breaks large, complex tables into semantically coherent sub-tables to improve reasoning efficiency and interpretability. This modular approach enables LLMs to handle large relational contexts, yielding significant performance improvements on benchmarks such as WikiTableQuestions and TabFact. Building on that, NormTab applied the principles of normalization to LLM-based reasoning. By systematically restructuring tabular inputs and reformatting cell values, NormTab demonstrated that structural clarity alone can significantly enhance symbolic reasoning and factual accuracy.
As my focus expanded, I explored how reasoning could be extended from static tables to dynamic, multi-source environments. This led to my current work, an agentic retrieval framework where multiple LLM-based agents collaborate to solve multi-hop question-answering tasks. In this work, specialized agents—such as an Analyzer, Selector, and Adder—interact iteratively to refine context, retrieve evidence, and validate answers. This multi-agent setup reflects a step toward self-organizing reasoning systems, capable of decomposing complex problems into coordinated subtasks. It showed measurable gains in precision and faithfulness across multi-hop datasets like HotpotQA and 2WikiMultiHopQA. Another challenge I am addressing lies in translating natural language into SQL queries for large, real-world databases. In this work, we propose a bidirectional retrieval method that integrates both entity-centric and attribute-centric linking strategies. Unlike traditional one-directional schema matching, this approach enables richer context understanding, improves schema recall, and reduces false positives. Tested on massive benchmarks such as BIRD and Spider, the framework demonstrates how retrieval-oriented reasoning can ground LLMs in large schema spaces efficiently.
Collectively, these projects chart a research trajectory that spans understanding, agentic retrieval, and reasoning—unified by a central ambition: to make LLMs not just generators of text but reasoning engines capable of grounded understanding, compositional logic, and reliable inference.
📆 News
- Attened COLM 2025 in Montreal, Canada
- ✨ Joined Huawei Canada as an Associate Researcher (Intern)
- 📢 Paper Alert! - Accepted one paper at EMNLP 2024
- 📢 Started PhD in Computing Science at UAlberta
- 📢 I have defended my MSc Thesis
- ✨ Attended NAACL 2024 in Mexico City
- 📢 Paper Alert! - Accepted one paper at NAACL 2024
- ✨ Attended AMII UpperBound Conference May, 2024
- ✨ Attended ACL 2023
Contact Me
✉️ Email
📧 Official: mnahid[at]ualberta[dot]ca
📧 Other: nahid-cse[at]sust[dot]edu
🔗 Website: mahadi-nahid.github.io
🔗 LinkedIn: mahadi-nahid
🐙 GitHub: mahadi-nahid
🐦 Twitter: mhn_nahid
🏠 Address
Department of Computing Science
1-13 Athabasca Hall
University of Alberta
Edmonton, Alberta
Canada T6G 2E8
—————————————-