TL Timothy Liu Kaihui

Timothy Liu

Applied AI researcher & solutions architect

I help research and engineering teams turn AI ideas into working systems. My work combines research, hands-on engineering, and technical guidance across models, GPU computing, and AI infrastructure.

Senior Solutions Architect, NVIDIA AI Technology Centre · PhD in NLP

Work Experience
LinkedIn GitHub Google Scholar
Portrait of Timothy

Selected projects

Engineering · Python · Apple Silicon

asitop

4.6k+ stars

I built a terminal performance monitor for Apple Silicon that brings CPU and GPU activity, memory, swap, and power into one view. Powered by macOS powermetrics and psutil, it also tracks Neural Engine power as an indicator of activity.

Recorded PyPI downloads
340k+
Later projects inspired by asitop
Explore asitop on GitHub →
asitop terminal display with CPU clusters, GPU utilisation, memory, Neural Engine power and power-history charts

12 later projects inspired by asitop

An influence beyond Apple Silicon

From independent Go, Swift, and Rust implementations to a commercial Mac interface and Linux/NVIDIA monitoring, asitop has influenced tools beyond its original Python implementation.

  1. mactop ↗

    Go monitor that credits asitop as its original inspiration.

  2. NeoAsitop ↗

    Independent Swift implementation that describes itself as asitop-inspired.

  3. pumas ↗

    A Rust reimplementation of asitop.

  4. FluidTop ↗

    An enhanced fork for newer Apple Silicon, terminal compatibility, and local-AI workloads.

  5. Mac Power Monitor ↗

    A commercial Mac app whose Apple Silicon summary window is designed to resemble asitop's display.

  6. asitop-csv-logger ↗

    A fork adding CSV logging for historical analysis.

  7. actop ↗

    Credits asitop as inspiration while using an independent architecture and codebase.

  8. DGXTOP ↗

    Extends the inspiration to Ubuntu and NVIDIA DGX Spark hardware monitoring.

  9. SiliconScope ↗

    Formerly WhisPlayInfo; names asitop and NeoAsitop among the motivations for its native SwiftUI dashboard.

  10. basitop ↗

    Credits asitop as a reference for powermetrics plist parsing.

  11. silicon-monitor / simonlib ↗

    Its Apple monitoring module documents asitop as its basis; the credit applies to that component.

  12. macmon ↗

    Acknowledges asitop as the original tool in its related-project list, establishing a predecessor relationship.
Media Coverage

Reviews and technical features

  • Apfeltalk · July 2022
    Michael Reimann's dedicated hands-on review explores performance analysis in the terminal.
  • Jeff Geerling · January 2025
    A dedicated section in his monitoring guide includes an asitop screenshot and installation instructions.
  • Genmind · May 2022
    Gian Paolo Santopaolo places asitop alongside nvitop in a practical AI-engineering toolkit.
  • davlgd labs · January 2024
    A French-language article examines Apple Silicon monitoring and power consumption.

International tutorials

Developer guides cover installation, troubleshooting, GPU activity, and energy monitoring across several languages.

Continuing recognition

Japanese publication AAPL Ch. continues to identify asitop when explaining later monitoring tools: the mactop launch (2024), macmon's M4 support (2024), mactop's menu-bar update (2026), and SiliconScope's release (2026).

Community use and recommendations

A Topaz Video AI community member shared screenshots of asitop while investigating video-upscaling workloads. Other discussions include local-LLM monitoring recommendations, Blender performance analysis, Hacker News recommendations, and MacRumors installation help.

Applied-AI collaboration

IEEE Access 2025 13 citations

FedDrip

Diffusion-generated images help federated learning models learn across hospitals with different data. As co-supervisor of a master's thesis project, I contributed research direction, ideation, and evaluation setup; the team developed and evaluated the method.

Read the FedDrip paper →

Findings of ACL 2023 29 citations

NewsMet

A news-headline dataset for interpreting and generating metaphors. I co-supervised the undergraduate capstone research project and advised as an NLP domain expert on dataset curation and benchmark development.

Read the NewsMet paper →

First-author research · ACL 2022 25 citations

Towards Better Characterization of Paraphrases

As first author, I introduced Word Position Deviation (WPD) and Lexical Deviation (LD) to measure changes in word order and vocabulary between paraphrases without expert annotation. The paper uses these metrics to evaluate datasets, improve model generalisation, and select paraphrases for data augmentation and robustness testing. Using MRPC and PAWS as case studies, I showed how the metrics reveal structural and lexical differences that conventional overlap scores can miss, while guiding dataset improvements and more controllable paraphrase generation.

I released paraphrase-metrics, an open-source Python library with almost 5k downloads. Hanyang University independently used WPD and LD to evaluate paraphrasing quality in its FCMR benchmark (published at ACL 2025), demonstrating external research adoption and validation of the metrics.

Read the paper → Code on GitHub

Experience

Research, architecture, and collaboration across Singapore and ASEAN/ANZ.

Aug 2026 - Present NVIDIA AI Technology Centre

Senior Solutions Architect

Support applied AI programmes with universities, research institutions, government, and industry across Singapore and ASEAN/ANZ.

  • Advise partners and NVAITC Joint Centres and Labs on research priorities and technical roadmaps.
  • Mentor teams through scoping, architecture, prototyping, and validation across foundation models, multimodal and agentic AI, GPU computing, simulation, and infrastructure.
  • Align regional initiatives through technical workshops and talks and centre launches and symposiums.
Aug 2024 - Aug 2026 NVIDIA AI Technology Centre

Solutions Architect

Supported applied AI programmes across ASEAN/ANZ, translating partner opportunities into research projects, prototypes, and roadmaps across generative AI and accelerated computing. Mentored research teams, contributed to publications, and co-organised workshops and talks and regional launches.

Sep 2023 - Dec 2023 Sony Research, Japan

Research Intern

Researched multimodal foundation models, paragraph-length captioning, and synthetic data for image-text retrieval with a Japanese R&D team. Fine-tuned 7B+ models on ABCI using multi-GPU training and DeepSpeed.

Jun 2018 - Jun 2023 NVIDIA AI Technology Centre

Solutions Architect Intern

Built ai-lab for reproducible experiments and supported Kubernetes GPU workloads on-premises and in the cloud during a five-year internship alongside my studies. Earned NVIDIA DLI Certified Instructor status for NLP, RAPIDS, and multi-GPU courses.

Selected research

Google Scholar Semantic Scholar

Google Scholar citation counts as of 17 Sep 2026.

ACL-IJCNLP 2021 7 citations Best Demonstration Runner-up Award

ESRA: Explainable Scientific Research Assistant

Hongwimol P., Kehasukcharoen P., Laohawarutchai P., Lertvittayakumjorn P., Ng A.B., Lai Z., Liu T., Vateekul P.

A literature search tool uses a knowledge graph to explain returned papers and show connections between research topics.

All publications (10)

The four selected papers above and the six additional publications below form the complete portfolio bibliography.

More open source

ai-lab

JavaScript

GPU-enabled AI development container for rapid prototyping and JupyterHub.

tf-metal-experiments

Jupyter Notebook

TensorFlow Metal experiments and model benchmarks on Apple Silicon.

prowler

Python

Raspberry Pi cluster-based distributed network vulnerability scanner.

Speaking & Events

Selected media coverage

About

My background spans NLP research, hands-on engineering, and technical collaboration.

Model development & evaluation

PyTorch, TensorFlow, HuggingFace Transformers, and DeepSpeed; paraphrasing, multimodal models, and synthetic data.

GPU systems & workflows

Docker, Kubernetes, Slurm, and PBS; reproducible experiments, GPU workloads, and multi-GPU training.

Applied-AI architecture & collaboration

Technical roadmaps, research mentorship, and prototyping with NVIDIA NIM, NeMo, and MONAI.

Education

Singapore University of Technology and Design

Sep 2020 - Aug 2024

PhD, Natural Language Processing

President's Graduate Fellowship scholar. Thesis: Foundations and Applications of Paraphrasing in NLP.

May 2017 - Aug 2020

Bachelor of Engineering (Honours)

Information Systems Technology and Design, with cybersecurity research and student workshops.

Complete toolkit
Languages
Python, SQL, Bash, HTML/CSS/JavaScript, Java
Frameworks
PyTorch, TensorFlow, HuggingFace Transformers, DeepSpeed, vLLM, SpaCy, NumPy, SciPy, OpenCV, XGBoost, Dask
Platforms
Linux/Unix, Git, Docker, Kubernetes, AWS, GCP, Azure, PBS, Slurm
Agentic coding
Codex, Claude Code, AWS Kiro, Cursor
NVIDIA technologies
GPU cluster deployment and management; NIM, Nemotron models, Blueprints, Brev, NeMo, RAPIDS, CuPy, Visual Search and Summarization (VSS), and MONAI

Spoken languages: English and Mandarin Chinese

Recognition and community work

Hyperhack Datathon

Gold Award for modelling medical-shock financial vulnerability with XGBoost and Tableau.

AI & Security Hackathons

Finalist at YITU Singapore AI Hackathon; winner at Google “Fake News Must Die” and Red Hack.

Black Hat Asia

Featured project for Prowler, a Raspberry Pi cluster network vulnerability scanner.

Community Teaching

Mentor at AI Saturdays Singapore and co-lead for Intel Make Tomorrow Catalyst workshops.

Other work

NSman

Mar 2023 - Present | Digital and Intelligence Service

Received a Letter of Commendation for Formation NSman of the Year and was invited to the Presidential Garden Reception 2026.

5 years of service SAF Good Service Medal

SpotterDex

Hobby project | Aircraft spotting and photography

A personal aircraft-spotting logbook and aviation photography field guide, with an interactive world map, aircraft index, squadron records, airshow collections, and spotting statistics.

Let’s connect

For research collaboration, technical exchange, or a speaking invitation, connect with me on LinkedIn.

Connect on LinkedIn GitHub Google Scholar