Portrait of Hongye Liu

Hongye Liu

PhD Candidate · Duke University

I work on NLP, reinforcement learning (RLHF / RLVR), LLM agents, model evaluation, and controllable generation. Advised by Ricardo Henao.

✉ Email 💻 GitHub 🎓 Scholar

About

I am a PhD candidate at Duke University (2024–2027, expected), advised by Ricardo Henao. My research centers on making large language models reason reliably over long, complex, and verifiable settings, from controllable generation and evaluation calibration to reinforcement learning with verifiable rewards (RLVR) and tool-augmented agents.

Alongside my studies I have been an Applied Scientist Intern at Amazon, and previously worked in the NLP group at Tencent AI Lab and on medical-NLP pipelines at BGI. My work has appeared at ACL, EMNLP and NAACL.

News

Selected Publications

* denotes equal contribution. Names are ordered as they appear in each work.

  1. Customer-Agent: Overcoming Context Limitations in Ultra-Long Shopping Trajectories via Tool-Augmented Agents and RLVR

    Hongye Liu, Rongmei Lin, et al.

    Under Review

  2. Learning to Control Summaries with Score Ranking

    Hongye Liu, Liang Ding, Ricardo Henao

    ACL Findings 2026

  3. Calibrating Evaluation Metrics for QA and Summarization

    Hongye Liu, Dhanajit Brahma, Ricardo Henao

    ACL Findings 2026

  4. Learning to Substitute Words with Model-based Score Ranking

    Hongye Liu, Ricardo Henao

    NAACL 2025 (Main)

  5. A Benchmark for Zero Pronoun Recovery and Translation

    Mingzhou Xu, Longyue Wang, Derek F. Wong, Hongye Liu, Linfeng Song, Lidia S. Chao, Shuming Shi, Zhaopeng Tu

    EMNLP 2022 (Main)

  6. Intelligent Flaw Detection of X-ray Images Based on Deep Learning

    Yang Shen, Hongye Liu, et al.

    Springer, 2021 · vol. 105, pp. 558–566

Patents (4)
  • A Method and Device of Bilingual Data for Web Fiction Translation via Self-Training (Hong Kong, 2021)
  • A Method and Device for Automatic Alignment and Automatic Translation of Web Fiction (Mainland China, 2021)
  • A Method and Device of Large-scale Data Filtering for Machine Translation (Mainland China, 2021)
  • A Ray Defect Detection Method based on the Mask R-CNN Model (Mainland China, 2018)

Experience

05/2026 – 08/2026

Applied Scientist Intern — Amazon

Palo Alto, CA · Manager: Xi Chen

  • Compressing memory into embeddings via model alignment.
  • Leveraging memory for downstream personalized question answering.
09/2025 – 12/2025

Applied Scientist Intern — Amazon

Seattle, WA · Managers: Rongmei Lin, Besnik Fetahu

  • Introduced a long-context shopping-QA benchmark (32k/64k) exposing LLM bottlenecks on long trajectories.
  • Proposed a Customer Agent Framework with verifiable SFT data synthesis and tool-based reasoning over externalized context files.
  • Demonstrated strong performance and generalization across internal and external reasoning benchmarks.
09/2020 – 09/2021

Research & Engineering Intern — Tencent AI Lab (NLP Group)

Shenzhen · Supervisors: Longyue Wang, Zhaopeng Tu

  • Web-novel translation: corpus alignment, back-translation, and NMT training pipelines; shipped a live translation engine.
  • Large-scale parallel-corpus filtering (N-gram + LM, Moore–Lewis selection) and zero-pronoun resolution test sets.
09/2019 – 01/2020

AI Algorithm Intern — Beijing Genomics Institute (BGI)

Shenzhen · Supervisor: Daoling Huang

  • Built medical-record NLP pipelines: word segmentation, NER, relation extraction, and knowledge-graph querying.
  • Containerized and deployed models on the biological gene-AI platform.

Awards