About
I’m a final‑year Ph.D. candidate in Computer Science at the University of Maryland, College Park in the Human‑Data Interaction Group, advised by Dr. Zhicheng Liu.
My work blends event sequence analysis, machine learning, and visual analytics: I build (1) benchmarks to evaluate multimodal models on structural properties of process visualizations, (2) source attributed RAG-based systems for trustworthy retrieval, and (3) human‑in‑the‑loop agentic systems that turn high‑level analytical intents into reliable workflows.
I have also developed domain-agnostic frameworks for event-sequence analysis and have conducted large-scale studies to understand how humans interpret information-rich event sequence visualizations.
Previously, I was an Applied Science Intern at AWS (Bedrock & MLSL), where I developed intervention-aware causal models for sequential data with large accuracy gains on industrial maintenance logs.
My work has been published at conferences and workshops of machine learning and HCI venues including EMNLP, IEEE VIS, EuroVis, CHI, NAACL and CVPR.
My dissertation, titled Beyond Domain Boundaries: Enhancing Generalizability of Event Sequence Visual Analytics Research, focuses on developing domain-agnostic modeling, analysis and visualization techniques for sequential data.
I hold a B.Sc. in Computer Science and Engineering from Bangladesh University of Engineering & Technology.
Outside of research, I enjoy reading (check out my Goodreads), traveling, hiking and exploring new cuisines.
News
Publications
Conference Papers
Workshop Papers & Posters
Software
Research
Vision-Language Model Evaluation for Process Visualizations
ProcVQA is the first benchmark for process visualizations: 118 charts with standardized rendering, 2,500+ ground‑truth tuples and 140+ expert VQA items. We evaluated 20+ models and exposed sharp drops from single‑ to multi‑hop reasoning plus higher hallucination on edge‑dense graphs.
Multi-Agent Event Sequence Analysis with LLM Orchestration
Operationalizes a task framework with specialized agents (Planner → Intent → Technique → Executor) in an approval‑gated loop using Autogen. Generates and runs analysis code with full traceability, combining hybrid retrieval (BM25 + vectors) and figure indexing.
Event‑Sequence Knowledge Base (RAG + Document AI)
ETL + retrieval pipeline that turns 100+ papers into structured, searchable knowledge. PDF parsing with figure/table extraction; multimodal indexing (text, captions, figure embeddings); hybrid retrieval (BM25 + vectors) with reranking; and source‑grounded answers delivered via a FastAPI backend and React frontend. Includes citation trails and confidence indicators to reduce hallucinations and support traceability.
Causal Inference in Temporal Event Sequences
Introduced intervention‑aware estimators within a temporal‑point‑process framework and a hybrid Transformer‑CNN architecture. Achieved 88% RMSE and 90% MAE reductions on industrial maintenance data and ~50% lower NLL in simulations; demonstrated literature‑consistent shifts on clinical data.
Multi-Level Multidimensional Event Sequence Analytics Framework
Systematic review of 58 systems across 16 venues yields a four‑level taxonomy (objectives → intents → strategies → techniques), with action–input–output–criteria quartets. Establishes a foundation for automation and knowledge transfer in sequence analytics.
Large-Scale Human-Centered Evaluation of Visual Summaries
180 participants × 6 datasets with 108 visual summaries. Reimplemented 3 pattern mining analysis techniques (CoreFlow, SentenTree, and Sequence Synopsis) with standardized encodings, revealing accuracy–efficiency trade‑offs (Sequence Synopsis highest quality; ~36% longer comprehension time).
COALA: Comparing Native and Non-native English Speakers in Collaborative Writing
Adapted Sequence Synopsis technique for behavioral pattern analysis, comparing native and non-native English speakers' writing strategies. Contributed to design implications for AI writing assistants through LLM-generated cluster summaries while maintaining interpretability.
Patterns of Inspiration: Analyzing Ideation Strategies in Example-Aided Visualization Design
Analyzed timestamped interaction logs of 32 visualization designers' example-aided ideation sessions. Developed comprehensive action classification framework categorizing behaviors into filtering, browsing, inspecting, curating, and sketching. Applied temporal pattern mining and statistical validation, revealing that timing of example introduction significantly influences filtering behaviors and that consistent anchoring with flexible filtering correlates with higher design quality and variety.
COVID-19's Unequal Toll: Small Business Impact Analysis
Assessed small business impact disparities with respect to ethnicity and race in US metropolitan areas using mobility data during COVID-19. Analyzed differential effects across communities and contributed to understanding of pandemic's unequal economic toll through data-driven approaches.
Interactive Bioinformatics Visualization Systems
Developed interactive visualization systems for single-cell RNA sequencing data with hierarchical annotations. Created scTreeViz R package enabling interactive exploration of complex biological datasets through icicle tree-based visualizations and data transformation components.
Experience
Applied Science Research Intern
Summer 2022, 2023Developed novel causal framework for detecting relationship shifts in temporal data. Worked with the Bedrock Generative AI Team (2023) and ML Solutions Lab (2022). Created hybrid Transformer-CNN architecture achieving significant performance improvements on real-world datasets.
Graduate Research Assistant
2021 - PresentAdvised by Dr. Zhicheng Liu. Research on event sequence analysis, VLM benchmarking, and multi-agent systems. Developed frameworks for event sequence analysis and created comprehensive evaluation protocols for vision-language models.
Teaching Assistant
2019 - PresentTeaching assistant for 7+ courses including Programming Languages (CMSC 430), Information Visualization (CMSC 471, CMSC 734), Bioinformatics (CMSC 423), Web Development (CMSC 335), and Data Science (CMSC 320).
Research Assistant
2019 - 2020Advised by Dr. Hector Corrada Bravo. Co-authored scTreeViz R package for interactive single-cell RNA sequencing visualization. Implemented data transformation components and visualization methods for hierarchical cell annotations.
Lecturer
2018 - 2019Instructor for Computer Networks, Structured Programming, Object Oriented Programming, Microprocessor, and Microcontrollers courses. Supervised undergraduate projects and research initiatives.
Awards & Recognition
Travel Grant
Department of Computer Science, University of Maryland
"Highly Useful" Reviewer Recognition
IEEE VIS, EuroVis, PacificVis
Jacob K. Goldhaber Travel Grant
Graduate School, University of Maryland
International Conference Student Support Award (ICSSA)
Graduate School, University of Maryland
AWS Cloud Credit for Research
$5,000 promotional credit for student research
Grace Hopper Conference Participation Grant
Iribe Initiative for Inclusion and Diversity in Computing (I4C), UMD
Summer Research Fellowship
College of Computer, Mathematical and Natural Sciences, UMD
Dean's Fellowship
Department of Computer Science, University of Maryland