Banghao ChiI am a M.S. CS student at the Siebel School of Computing and Data Science, part of the Grainger College of Engineering at University of Illinois at Urbana-Champaign, advised by Prof. Minjia Zhang, working on LLMs inference optimization. At the National Center for Supercomputing Applications (NCSA), I worked with Dr. Kevin Chang on developing advanced information retrieval systems using LLMs. Prior to that, I studied Mathematics at UIUC and Computer Science at Xi'an Jiaotong Liverpool University. During my time there, I worked with Prof. Xiaohui Zhu on developing LLMarking, an innovative automatic short answer grading system. Email / GitHub / Google Scholar / LinkedIn / CV |
|
PublicationsI'm interested in LLMs inference optimization and its applications. |
|
LLMarking: Adaptive Automatic Short-Answer Grading Using Large Language ModelsHanling Wang, Banghao Chi, Yufei Wu, et. al. Association for Computing Machinery Learning@Scale (ACM L@S), 2025 paper / code / poster / We present LLMarking, a novel automatic short-answer grading system that leverages large language models along with a Key Point Scoring Framework and Prompt Dynamic Adjustment to deliver flexible, accurate, and explainable assessments across various educational contexts. |
|
Research Advanced in the Object Detection Based on Deep LearningBanghao Chi International Conference on Applied Physics and Computing (ICAPC), 2022 paper / We provide a comprehensive review and experimental analysis of recent advances in object detection based on deep learning, categorizing methods into traditional, anchor-based, and anchor-free frameworks, and highlighting their performance and future challenges. |
Other ProjectsThese include coursework, side projects and unpublished research work. |
|
LLMs-based Knowledge AgentsNCSA-SPIN 2025-05-16 paper / video / code / We present a structured, modular information retrieval system that combines Finite State Machines (FSMs) with Large Language Models (LLMs) to automatically extract and enhance entity-specific information from the web, using recursive link analysis, dynamic schema generation, and JSON-based structured outputs. |
|
Q-LiDAR: Efficient and Accurate Training-Free Quantization for Point Cloud 3D Object Detection ModelsUIUC-SSAIL 2025-01-01 paper / code / We introduce Q-LiDAR, a training-free quantization framework for 3D LiDAR object detection models that improves inference efficiency without compromising accuracy by combining component-specific techniques like SmoothQConv, channel-wise quantization, and Hessian-guided bit-width allocation. |
Working Experience |
Awards & Honors |
| Year | Award | Institution |
|---|---|---|
| 2025 | University Dean's List (top 20% excellence) | U of I at Urbana-Champaign |
| 2024 | University Dean's List (top 20% excellence) | U of I at Urbana-Champaign |
| 2023 | University Academic Excellence Award (top 1% excellence) | Xi'an Jiaotong Liverpool University |
| 2022 | Summer Undergraduate Research Fellow | Xi'an Jiaotong Liverpool University |
| 2022 | University Academic Achievement Award (top 2% excellence) | Xi'an Jiaotong Liverpool University |
| 2022 | Awarded 2nd Prize in Asia and Pacific Mathematical Contest in Modeling | Consortium for MAP |
| 2021 | Awarded 2nd Prize of FLTRP Cup National English Speaking Contest | Foreign Language Research Press |