Sepideh Abedini
Office: DC 3301
University of Waterloo
Waterloo, ON N2L 3G1 Canada
My name is Sepideh Abedini. I am an Applied ML Specialist at the Vector Institute and recently completed my MMath in Computer Science at the University of Waterloo, where I was supervised by Prof. M. Tamer Özsu in the Data Systems Group.
My research lies at the intersection of Large Language Models, Data Management Systems, and Privacy-Preserving Machine Learning. I focus on developing reliable and privacy-aware LLM systems for structured reasoning and enterprise applications, with particular emphasis on Text-to-SQL, context-aware PII detection, benchmarking, and privacy-enhancing technologies for LLM pipelines.
During my time at Vector Institute, I led projects on privacy-preserving Text-to-SQL and context-aware PII detection for downstream QA systems. My recent works include MaskSQL, a privacy-preserving Text-to-SQL framework accepted at the NeurIPS 2025 Regulatable ML Workshop, and CAPID, a context-aware PII detection framework accepted at EACL 2026 SRW. I also developed SQLyzr, a comprehensive benchmark and evaluation framework for fine-grained analysis of Text-to-SQL systems.
In my free time, I enjoy socializing and engaging in conversations with people to learn new perspectives.
__________________________________________________________________________________
Experiences
-
Applied ML Specialist / ML Research Student - Vector Institute
- Proposing and leading research on privacy-aware LLM systems, with a focus on privacy-preserving Text-to-SQL and context-aware PII detection.
- Developed MaskSQL, a framework that abstracts sensitive schema and value tokens before LLM inference under customizable privacy policies.
- Collaborated on designing a context-aware PII detection pipeline combining NER, rule-based filtering, and SLMs to selectively mask sensitive spans while preserving downstream utility.
- Collaborated on building synthetic data generation pipeline for context-aware PII benchmarking, resulting in the creation of more than 2K examples for CAPID.
-
Research Assistant - University of Waterloo
- Developed SQLyzr, a comprehensive benchmark and evaluation framework for fine-grained analysis of Text-to-SQL systems.
- Designed taxonomy-driven evaluation methodologies for analyzing correctness, complexity, efficiency, and robustness of NL-to-SQL models.
- Conducted large-scale benchmarking and workload alignment studies on state-of-the-art Text-to-SQL systems.
-
Teaching Assistant - University of Waterloo & Sharif University of Technology
- Teaching Assistant for several undergraduate computer science courses covering functional programming, systems programming, algorithms, and introductory software development.
-
Machine Learning Engineer Intern - JobVision
- Developed an AI recommender system for job-seekers
- Implementing content-based and collaborative filtering algorithms
Education
-
MMath in Computer Science, University of Waterloo
- Thesis: SQLyzr: A Comprehensive Benchmark and Framework for Evaluating Text-to-SQL Systems
- B.S. in Computer Science, Sharif University of Technology
Honors & Awards
- Awarded the International Master’s Award of Excellence (IMAE), University of Waterloo
- Awarded Full Scholarship based on National Universities Entrance Exam, Sharif University of Technology