LLM Hallucination Detection
Developing methods that combine model uncertainty, entropy-guided probing, and knowledge graph features to predict when LLM responses are likely to be factually unreliable.
About
Researching more reliable large language models
I am a PhD candidate in Computer Science at the Knowledge and Data Engineering (KDE) Lab, University of Tsukuba, advised by Prof. Toshiyuki Amagasa. My research focuses on improving the factual reliability of large language models through structured knowledge, particularly knowledge graphs. I work on knowledge graph-LLMs alignment for solving real-world problems.
Alongside my doctoral research, I work as a Research Assistant at National Institute of Advanced Industrial Science and Technology (AIST), Japan. My research also extends to broader problems at the intersection of artificial intelligence and data management, where I explore how emerging AI methods can be applied across different data types, domains, and real-world scientific problems.
Large Language Models
Knowledge Graphs
Agentic AI
Trustworthy AI
Graph Reasoning
Hallucination Detection
I am currently exploring postdoctoral research opportunities related to
large language models, knowledge graphs, trustworthy AI, agentic AI and knowledge-enhanced reasoning.
Research
Current research directions
My work centers on using structured knowledge to detect, understand, and reduce factual errors in large language models.
Knowledge Graph Augmentation
Designing adapter-based and selectively gated mechanisms for injecting structured knowledge into frozen LLMs while avoiding unnecessary knowledge augmentation.
Agentic AI
Exploring agentic approaches for integrating knowledge graphs with LLMs to improve reasoning and knowledge utilization.
Publications
Selected publications
Under review / submitted
When to Inject: Gated Adapter-Based Selective Knowledge Graph Augmentation for Large Language Models
Knowledge Graph-Guided Multi-Hop Probing for Hallucination Detection in Large Language Models
LARA-Event: Lexicon-Audited Robust Adaptive Fusion for Event-Centric Opinion Mining
2026
When Structure Predicts Hallucination: Aligning LLMs with Knowledge Graph Features
Knowledge Graph Adapter-Based Augmentation Testbed for Large Language Models
Risk-Aware KG Adapter for Improving LLM Accuracy
2025
Entropy-Guided Probing for Predicting LLM Hallucinations with Knowledge Graph Features
Multi-Hop Corpus for Detecting LLM Hallucinations
Academic background
Education
Apr 2024 — Expected Mar 2027
Ph.D. in Computer Science
University of Tsukuba, Japan
Supervisor: Prof. Toshiyuki Amagasa. Research focus: aligning large language models with knowledge graphs for improved factuality and reliability.
Oct 2020 — Aug 2023
M.Phil. in Computer Science
University of Central Punjab, Lahore, Pakistan
Thesis: Opinion Mining of Socio-Political Tweets Using BERT Embeddings.
2016 — 2020
B.S. (Hons) in Computer Science
PMAS Arid Agricultural University, Rawalpindi, Pakistan
Experience
Research & Professional experience
Jul 2024 — Present
Research Assistant : AIST, Japan
Conducting research on graph-based methods for large language models, with emphasis on factual reliability, hallucination detection, structured knowledge integration, and model evaluation.
Apr 2026 — Present
Research Assistant : University of Tsukuba, Japan
Supporting research on knowledge graph and LLM integration, experimental evaluation, and academic publication.
Sep 2024 — Present
Teaching Assistant : University of Tsukuba, Japan
Supporting research seminars and student learning activities.
Aug 2022 — Mar 2024
Data Analyst : eConceptions, Islamabad, Pakistan
Developed Python-based data analysis and predictive modeling workflows, managed MySQL data pipelines, and built operational dashboards and reports.
Mar 2022 — Aug 2022
Lab Instructor : University of Central Punjab, Lahore, Pakistan
Taught introductory computing and C++ programming laboratory sessions.
Recognition
Awards & scholarships
-
Apr 2024 — Present
MEXT Scholarship
Ministry of Education, Culture, Sports, Science and Technology, Japan.
-
M.Phil.
Merit Scholarship
University of Central Punjab, Lahore, Pakistan.
Contact
Get in touch
I am happy to discuss research collaborations, related work, and research opportunities. Email is the best way to reach me.
Tsukuba, Ibaraki, Japan