Zhengyang (David) Li
AI Software Engineer | Applied AI, LLM Systems & MLOps
I build production-oriented AI systems that connect multimodal data, model inference, embeddings, vector search, asynchronous services and APIs. I am based in Adelaide, fluent in English and Mandarin, and hold full Australian work rights.
About
I currently work as a Software Engineer at Adelaide University and a Full Stack AIOps Engineer at SoMe AI. My work spans Python services, multimodal processing, LLM integration, vector databases, GPU inference, containerised deployment and production observability.
My background crosses engineering, data science, teaching and research. I enjoy the part of AI work that happens after a model demo: designing reliable workflows, building the surrounding software, and making the resulting system observable and maintainable.
This site is both a portfolio and a working notebook. I write about AI systems, Linux infrastructure, Python, self-hosting, experiments and the occasional personal reflection. Older articles remain as a record of how my interests and practice have evolved.
Open-source spotlight
AI Usage Dashboard
A privacy-first browser extension that consolidates source-visible quotas, spending, reset times and sync health for Codex, Claude, Cursor, Gemini Code Assist and custom APIs across toolbar, side-panel and full-page views.
I built the provider adapters and normalised state handling in TypeScript and React, keeping credentials and cached data in the user's browser. The project includes responsive themes, 14-language and RTL support, Vitest and Playwright QA, and GitHub Actions release automation.
Technology: TypeScript, React, Vite, WebExtensions, Chrome APIs, Vitest, Playwright and GitHub Actions.
What I work on
Applied AI & multimodal systems
Video, image and text preprocessing; model inference; post-processing; embeddings; retrieval; API delivery; and asynchronous workflow orchestration.
LLM systems & MLOps
Hosted and local LLM APIs, MCP and agent components, GPU inference, containerised deployment, evaluation and operational monitoring.
Product engineering
TypeScript and React applications, browser extensions, REST APIs, data stores, automated testing and CI/CD release pipelines.
Selected experience
Software Engineer — Adelaide University
Jul 2026–Present · Contract
- Design and implement end-to-end AI processing workflows covering multimodal preprocessing, inference, post-processing, embeddings, storage, search and retrieval, API delivery and asynchronous orchestration.
- Develop Python, SQL and Bash services integrating PyTorch and Hugging Face model stacks, hosted and local LLM APIs, MCP and agent components, vector databases and supporting data stores.
- Deploy containerised GPU workloads on Linux and monitor GPU utilisation, VRAM, latency, throughput, API errors, queue depth and system load.
Full Stack AIOps Engineer — SoMe AI
Jun 2025–Present · Part-time · Promoted from AI Software Engineer Intern
- Build multimodal video and audio processing workflows, including ASR, embeddings, product matching, vector retrieval and evaluation tooling.
- Develop and deploy Dockerised Python and REST services on multi-GPU Linux systems for media processing, retrieval and production analysis.
Teaching Assistant / Tutor — University of Adelaide
Aug 2023–May 2025 · Part-time
- Facilitated workshops and drop-in support for master's-level Python students, with small-group and one-to-one debugging guidance.
- Improved slides, code examples and lab materials in collaboration with the teaching team.
DevOps Engineer — Hygon
Jul 2020–Aug 2021 · Full-time
- Developed high-concurrency Python services using asyncio and automated Linux operations with Bash and Python.
- Supported and optimised high-performance computing environments.
Research & teaching
I have published three computer-science papers on time-series classification and model robustness, including two as first author, and received an ADC Best Student Paper Award. I also spent nearly two years supporting master's-level Python teaching at the University of Adelaide.
- Robustness Analysis on Self-Ensemble Models in Time Series Classification — ADMA 2024
- Boosting Certificate Robustness for Time Series Classification with Efficient Self-Ensemble — CIKM 2024
- Evaluating Model Robustness Using Adaptive Sparse L0 Regularization — ADMA 2024
Education
- Master of Data Science — University of Adelaide, completed Jul 2025, GPA 6.33/7
- Master of Physics — Soochow University, 2020
- Bachelor of Engineering — Shenyang Ligong University, 2016
Technical toolkit
Languages: Python, TypeScript, SQL and Bash/Shell
AI & multimodal: PyTorch, Transformers, Hugging Face, vLLM, Ollama, TensorRT, ONNX Runtime, OpenCV, FFmpeg, Whisper and SAM
LLM & agents: OpenAI API, Anthropic API, Gemini API, MCP and agent frameworks
Applications, data & platforms: React, Vite, REST APIs, WebExtensions, PostgreSQL, Supabase, Redis, MongoDB, Qdrant, Linux, Docker/Compose, Nginx, GitHub Actions, GitLab CI/CD, Proxmox and VMware ESXi
Contact
For professional enquiries, open-source collaboration or a conversation about applied AI systems, email [email protected] or connect with me on LinkedIn.
