Albert Yu Sun

Publications

InvestigationWorlds: An Agentic Environment for Legal Investigation

A Sun, A Benard, S Hamilton, A Marcelo, Y Kim, C Henneking, R Hu, Y Wang, D Mimno, B Yang, I Labutov

NeurIPS Main Conference, Evaluations and Datasets Track 2026

Paper
CorporateBench: Large-Scale Q&A Benchmarking with Temporal Knowledge Bases

S Hamilton, A Sun, O Romero, C Henneking, D Mimno, B Yang, I Labutov

EMNLP Findings 2026.

Paper Website Dataset

🏆 Featured in Fireworks’ Specialized Intelligence Index for model evaluation.
CONSCENDI: A Contrastive and Scenario-Guided Distillation Approach to Guardrail Models for Virtual Assistants

A Sun, V Nair, E Schumacher, A Kannan

NAACL 2024. Granted US Patent 12,259,952 B2.

Paper Dataset Patent
Does Fine-Tuning GPT-3 with the OpenAI API Leak Personally-Identifiable Information?

A Sun, E Zemour, A Saxena, U Vaidyanathan, E Lin, C Lau, V Mugunthan

arXiv 2023. 35+ citations.

Paper
Navigating progress: Enhancing public transit for more equitable communities via interpretable causal inference

G Parikh*, J Huang*, A Sun*, L Semenova, C Rudin

Harvard Data Science Review 2025.

Paper

Also presented at: NeurIPS 2022 Causal Machine Learning for Real-World Impact

🏆 Best Student Paper Award, 2022 American Statistical Association Data Challenge Expo Student Competition
Multi-Analyst Differential Privacy for Online Query Answering

D Pujol, A Sun, B Fain, A Machanavajjhala

VLDB 2022.

Paper

Also presented at: ICML 2022 Theory and Practice of Differential Privacy Workshop

Experience

Epiq AI Labs - AI Research Engineer (2025-present)
Dynamo AI - ML Research Engineer II (2023-2025)
Curai Health - Research Intern (2023)
Microsoft Research - Research Intern (2022)
Vera Institute of Justice - Data Science Intern (2020-2021)
ACLU - Legal Intern (2020)

Research News

September 2026: Our paper, "InvestigationWorlds: An Agentic Environment for Legal Investigation", was accepted as a poster to the NeurIPS 2026 Evaluations and Datasets Track!
August 2026: Our paper, "CorporateBench: Large-Scale Q&A Benchmarking with Temporal Knowledge Bases", was accepted to EMNLP Findings 2026! See you in Budapest.
March 2024: My first-author paper, "CONSCENDI: A Contrastive and Scenario-Guided Distillation Approach to Guardrail Models for Virtual Assistants", was accepted to NAACL 2024! See you in Mexico City.
March 2023: I served as a reviewer for EMNLP 2023.
July 2023: We submitted our preprint "Does fine-tuning GPT-3 with the OpenAI API leak personally-identifiable information?" to Arxiv. This was my first project at DynamoFL, and we were first to analyze privacy leakage from OpenAI's fine-tuning API.