Experience
CMB Wing Lung Bank
Aug 2026 — Present · Hong Kong
Assistant Manager — Wing Lung AI Hub: Enterprise AI Agent Platform
- Led product planning and iteration for Wing Lung AI Hub; repositioned it from an AI tool store into the bank's central hub for AI adoption, owning the lifecycle from requirements to launch.
- Build AI agent workflows on an enterprise agent platform to automate manual business processes, extending the low-code platform with custom Python components for file handling and data processing.
Huawei Hong Kong Research Center
Sep 2024 — Aug 2026 · Hong Kong
R&D Engineer (Research Intern, Apr — Jul 2024)
AI for RTL — CLI Agent Skills for Document Intelligence
- Built CLI-based agent skills that convert engineering PDF / Word / Excel documents into Markdown chunks and hierarchical indexes via MinerU and PageIndex; reduced MinerU parsing time from ~2 min to ~30 s and parallelised index-summary generation with a lightweight 4B model.
- Designed human-in-the-loop checkpoints that let users validate results step by step, bringing end-to-end processing of a 20+ page document down to ~4 min.
- Deployed LLMs on Linux servers with multiple graphics cards using Docker and vLLM, unified behind a LiteLLM gateway with an OpenAI-compatible API for agent development.
Intelligent Circuit Review Engine — Code Generation System
- Architected a multi-agent workflow (Writer / Debugger / Critic) in LangGraph with a sandboxed generate-execute-critique loop; achieved an 80%+ script pass rate and shortened task time from 0.5–1 day to minutes.
- Served Qwen models on Huawei Ascend NPUs behind a dynamic load balancer, scheduling requests across concurrent instances to maximise throughput under multi-tenant agent workloads.
- Exposed the system via RESTful APIs and MCP services, and built an asynchronous Redis-queue I/O pipeline so multiple agents stream responses concurrently without blocking.
Hardware Design Document Understanding
- Built an end-to-end pipeline from scratch that turns unstructured PDFs into structured data, combining deep-learning layout analysis and table recognition with LLM post-OCR correction; achieved a 90% benchmark pass rate.
- Deployed the model stack as a Model-as-a-Service (MaaS) layer on Huawei Cloud ModelArts, exposing a unified API for downstream applications.
Centre for Artificial Intelligence & Robotics, CAS
Dec 2023 — Apr 2024 · Hong Kong
Intern — Multimodal Large Model for Healthcare
- Adapted general-purpose 7B / 14B LLMs into a medical-domain assistant via continual pre-training, SFT and prompt engineering on medical corpora; achieved 80.4% accuracy on the PubMedQA benchmark.
- Built a RAG system over medical knowledge sources with Milvus and ChromaDB, including text-corpus construction and embedding pipeline design for retrieval.
- Ran distributed fine-tuning of 7B / 14B models on an NVIDIA A800 GPU cluster via LLaMA-Factory.
Skills
Programming & Tools
Python (expert), SQL, C++ (basic), Bash, Linux, Git
LLM & Agents
LangGraph, MCP, RAG, Tool Calling, Claude Code, Dify, n8n, MinerU, PageIndex
Serving & Data
vLLM, SGLang, LiteLLM, Ollama, FastAPI, Docker, Redis, Milvus, ChromaDB
Education
MSc in Computer Science (with Credit) — City University of Hong Kong
Sep 2023 — Oct 2024
BEng in Computer Science and Technology — South China Normal University
Sep 2019 — Jun 2023
Award: Meritorious Winner, Interdisciplinary Contest in Modeling (ICM), 2022.
Contact
I am currently exploring AI engineer opportunities in Hong Kong. You can reach me on LinkedIn.