Harry Y. H. Li | 李裕康
About
I'm interested in a broad range of topics in computer science. My research interests focus on TCS B, particularly programming languages, as well as software engineering and systems. I also have experience in full-stack software engineering, data mining, analysis, and visualization.
Education
Awards & Achievements
Talent Development Scholarship
2025-26HKSAR Government Scholarship Fund
Reaching Out Award
2025-26HKSAR Government Scholarship Fund
Dean's List
2025-26 FallSchool of Engineering, HKUST
IET Prize 2025
2025-26IET Hong Kong
Fung Scholarships
2025-26Victor and William Fung Foundation
Best UGTA Award
2024-25Department of Computer Science and Engineering, HKUST
AEON Credit Service (Asia) Scholarships
2024-25, 2025-26Hong Kong University of Science and Technology
Mr & Mrs Poon Kuen Seng Memorial Scholarships
2024-25, 2025-26Department of Computer Science and Engineering, HKUST
Dean's List
2024-25 Fall, SpringSchool of Engineering, HKUST
University's Scholarship Scheme for Continuing Undergraduate Students
2024-25, 2025-26Hong Kong University of Science and Technology
HKUST Alumni Endowment Fund High Flyers Program Scholarship
2023-24Hong Kong University of Science and Technology
Best UGTA Award - Honorable Mention
2023-24Department of Computer Science and Engineering, HKUST
Dean's List
2023-24 Fall, SpringSchool of Engineering, HKUST
University Admissions Scholarship
2023Hong Kong University of Science and Technology
Research Experiences
Undergraduate ThesisOngoing
2026 - 2027Flow-Based Program Optimization
COMP 4981H · Project LP4. Advised by Prof. Lionel Parreaux.
I'm investigating how reusable flow analysis can support program optimizations that preserve behavior in MLscript, TACO Lab's experimental programming language. The project uses flow webs to group related producers and consumers, such as functions and their call sites, so that transformations can update them together.
Eta expansion can move a computation into a lambda, changing when or how often it runs, including when it mutates or observes a state. I implemented an effect-analysis prototype and integrated it into the existing eta-expansion optimization, avoiding transformations that may move observable effects or state mutations and observations. The analysis propagates effects through calls, including recursive calls and calls to imported functions.
Other directions I'm considering include storing reusable analysis results in symbols, extending flow webs to constructors and destructors, preserving web relationships during deforestation, generalizing dead-parameter elimination to dead-everything elimination, and investigating further transformations such as defunctionalization.
Research CollaborationOngoing
Since 2025-26 SpringOpportunistic Parallelism
I work on PopPy, a compiler for parallelizing Python agentic programs, at the University of Pennsylvania with Prof. Osbert Bastani and Dr. Stephen Mell. I joined the collaboration in Spring 2026 and also took part in a summer research internship.
I extended Bezoar, PopPy's intermediate language, with control transfer statements (early return, break, and continue) and a general for/while loop construct. I formalized the extension's syntax, semantics, and lowering to core Bezoar.
I also worked on control-flow transformations, iterator sequencing, and memory scheduling, and adapted agent-workflow benchmarks for evaluation.
I'm designing a unified effect system to broaden PopPy's support for effects such as object mutation and I/O. This ongoing work connects with my thesis on flow-based program optimization.
PopPy: Opportunistically Exploiting Parallelism in Python Agentic Workflows
Stephen Mell, David Mell, Harry Y. H. Li, Zhihao Jia, Konstantinos Kallas, Steve Zdancewic, and Osbert Bastani
arXiv preprint and submitted to ASPLOS 2027 (under review)
Undergraduate Research Opportunities Program
2024-25 Fall, Spring, Summer. 2025-26 FallProgramming Language Design and Implementation
Advised by Prof. Lionel Parreaux, I investigated how to support ad hoc polymorphism in MLscript, our lab's experimental programming language, while keeping compilation independent of full type inference. The challenge was to recover enough information at each use site to select the intended definition and supply implicit arguments.
- Module System. I first implemented a restricted module-based approach that kept callable declarations known at compile time. Checks on module-valued parameters, return values, and bindings preserved the information needed to insert implicit arguments as modules passed through a program. I also developed formal module-checking rules to describe these restrictions.
- Resolution Pass. Forward references and chained member accesses could require declarations that elaboration had not yet processed. I developed a separate pass after elaboration, combining symbol resolution with lightweight type facts from annotations, literals, constructors, and known members. It also disambiguates overloaded names, such as a class and companion module sharing a name, and preserves the selected definition through lowering and JavaScript generation.
- Type Classes and Contextual Parameters. Building on this information, I implemented lookup and insertion of matching instances from lexical scope, letting functions receive implementations implicitly. I added constructor support, explicit contextual arguments, and instance retrieval with
use[T]. Alongside the implementation, I formalized the translation into explicit calls and explored a declarative algorithm for interacting instance requirements; the general algorithm remained unimplemented. - Multiple Parameter Lists and Eta Expansion. To support ordinary and contextual arguments together, I added multiple parameter lists across parsing, elaboration, and code generation, enabling partial application. I then implemented eta expansion so functions with interleaved parameter lists could be used as values with implicit arguments supplied, and refined it to avoid redundant expansion inside application chains.
- Control-Flow Analysis. The module-based approach restricted how values could be passed around. I explored using flows from producers to consumers to recover information beyond local annotations and prototyped basic implicit-argument resolution on this basis. I found simple 0-CFA sufficient for the intended ad hoc-polymorphism use cases in MLscript. This investigation led to my undergraduate thesis on using flow information for program optimization.
Undergraduate Research Opportunities Program (Tasting Stream)
2023-24 SummerCommonsense Reasoning with Knowledge Graphs
I was advised by Prof. Yangqiu Song and PhD candidate Zihao Wang. I studied a postgraduate course about graph machine learning, graph neural networks, and knowledge graphs. I participated in a research project about logic reasoning over knowledge graphs, where I gained hands-on experience in the research field.
Work Experiences
Independent Work Project
2026-27 FallIntelligent Feedback and Style Coach for Programmers
Advised by Prof. Desmond Tsoi.
- Developed a TypeScript VS Code extension and browser prototype that gives novice Python programmers code-local feedback on correctness, clarity, and performance, with explanations and student-controlled edit previews.
- Evaluated the prototype on a development benchmark of 450 course-derived and multi-issue inputs with paired controls, using LLM adjudication to assess feedback.
Independent Work Project
2025-26 Fall, WinterCRS Request System
Advised by Prof. Desmond Tsoi. The system serves COMP 1023, COMP 2011, COMP 2012, and COMP 2211, with over 2,000 active users.
- Built student and instructor workflows for class swaps, absences, and assignment deadline extensions.
- Designed the application with TypeScript, Zod, tRPC, MongoDB, and React/Next.js.
- Mentored subsequent contributors as the project continued to develop.
Independent Work Project
2024-25 Spring, 2025-26 FallAI Agents as Dynamic, Adaptive Language Assessors: Building a Patentable App
Co-advised by Prof. Desmond Tsoi and Prof. Mei Yi Angel Lin at the Education University of Hong Kong.
- Developed an app that uses multimodal language models to help learners practice English and assess their language skills.
- Supported Mandarin- and Cantonese-speaking learners with translation assistance during English conversations.
Department of Computer Science, HKUST
2023-24 Spring - NowUndergraduate Teaching Assistant
As Main TA for COMP 1023, I teach the course's tutorial sections. In other UGTA roles, I help deliver tutorials and labs, prepare course materials, and answer students' questions. My winter and summer appointments focus primarily on preparing assignments, tutorial notes, lecture notes, and self-tests.
- COMP 1023 - Introduction to Python Programming
Course coordinator: Prof. Desmond Tsoi.
Main TA: Fall 2026, Fall 2025; UGTA: Summer 2025, Summer 2024. - COMP 2011 - C++ Programming
Course coordinator: Prof. Brian Mak.
UGTA: Spring 2025, Winter 2025, Spring 2024. - COMP 2012 - Object-Oriented Programming and Data Structures
Course coordinator: Prof. Brian Mak.
UGTA: Winter 2026, Fall 2024.
GREAT Smart Cities Institute, HKUST
Summer 2024 - approx. Fall 2025Summer Internship
- Built computer-vision tools to automate traffic-data processing and analysis.
- Integrated the results with GIS software for visualization and spatial analysis.
- Analyzed traffic patterns and improved the team's data-processing workflows.
USThingCore MemberVoluntary
2023-24 Fall - NowHead of Backend & App Team Leader
USThing is a student-run app for HKUST. I led the backend and app teams within the 26-member organization.
I led a rewrite from June to August 2024 to bring an unmaintained backend and its dependent features back into service.
- Rebuilt the backend around containerized microservices and migrated separate native Android/iOS applications to React Native.
- Used TypeScript types and OpenAPI contracts to keep the frontend and backend interfaces consistent.
- Coordinated development across both teams and improved CI/CD and release workflows.
The app later recorded 15,000 active users over a 90-day period and up to 5,000 daily active users.
Other Experiences
Student Representative
2026-27 Residential YearHall Affairs and Operations Committee (HAOC), HKUST
Nominated by Prof. Cindy Li, I am serving as a student representative on the Hall Affairs and Operations Committee, contributing a student perspective to the review of hall-residence appeals and the interpretation of hall rules.
Artifact Evaluation Committee
2026ICFP 2026, Indianapolis, IN, USA
Served on the ICFP 2026 Artifact Evaluation Committee (co-chaired by Son Ho and Prof. Lionel Parreaux), reviewing artifacts submitted to the International Conference on Functional Programming.
AI Literacy Initiative WorkshopPreparing
2026-27Hall X, HKUST
Working with Prof. Cindy Li and fellow students on AI literacy for Hall X residents, including first-year students. I am developing two beginner-friendly, 90-minute workshops combining live demonstrations and hands-on activities.
- Introduction to AI and Agents: A Practical Starter Kit. A structured introduction to AI providers, models, chatbots, and agents, with practice in specifying tasks, supplying context, and checking responses. Participants will leave with a reusable prompting template and verification checklist.
- Everyday Agentic Workflows: From Chatting to Getting Things Done. A practical introduction to using agents for multi-step student tasks, with clear inputs, outputs, permissions, and review points. A guided workflow will cover gathering information, preparing documents, protecting personal information, and reviewing consequential actions. No programming background is required.
University of Oregon
2025-26 SummerOregon Programming Languages Summer School
Attended the Oregon Programming Languages Summer School (OPLSS), a two-week intensive program on programming language theory. The topic was Types, Proofs, and Program Logics.
UG Student Representative
12 May 202532nd Meeting, Student-Staff Liaison Committee, CSE, HKUST
Skills
Programming Languages
Libraries / Frameworks / DBMS
Tools
Projects
My personal website and portfolio, covering research, projects, teaching, and other experiences.
A platform for students and instructors to manage class swaps, absences, and deadline extensions. Used across four HKUST courses, with over 2,000 active users.
An experimental to-do language and VS Code extension, with a compiler written in Rust. It supports relative dates and times, with a hand-written parser and pretty-printer.
Friends
UG in ... IDK @ HKUST '29
Funny science student who claims to be from the business school 🐽
UG in DS. @ HKUST '27
Exchange @ ETH Zurich
Research Interest:
He doesn't know either. 😛
UG in CS+Math @ HKUST '25
PhD in CS+Stats @ UPenn
Theory and Empirical Science of Deep Learning and LLMs 🥺