👋 Hi, I’m Kuo Wei
Staff-level backend engineer with 10+ years of experience in distributed systems, and high-throughput AI platforms. Expert in designing scalable GenAI pipelines, multimodal LLM integration, and large-scale data platforms, driving innovations that power millions of users.
| 📫 Connect with me on LinkedIn |
💼 Professional Experience
TikTok · Singapore · 2023 – Present
- Drove 15% of total daily new users (>10K DNU) by architecting a high-throughput GenAI pipeline, orchestrating text-to-text and text-to-image models to auto-generate ad copy, images, and videos, achieving 5× faster creative production.
- Built a multimodal creative insights platform analyzing 100K+ ad assets (competitor ads, in-house campaigns, and TikTok UGC videos), delivering data-driven insights that guided large-scale advertising strategy.
- Developed an AI-powered creative agent with chat-driven exploration, bulk generation, and vector-based retrieval, streamlining workflows across internal and external teams.
Alibaba · Hangzhou, China · 2017 – 2023
Security Innovation Lab (SIL), Alibaba Cloud
- Architected a cloud-scale cyber risk quantification platform, leveraging large-scale ML + LLMs to detect emerging threats and automate high-throughput security workflows, protecting millions of enterprises on Alibaba Cloud.
- Led a redesign for multi-region, fault-tolerant deployments, achieving 99.99% availability while ensuring regulatory compliance and real-time threat detection.
- Automated infra provisioning via Terraform + serverless computing, reducing costs by 10× for bursty workloads.
Machine Intelligence Technology Dept. (MIT), DAMO Academy
- AI Work Assistant: Architected and led a scalable AI-powered work assistant integrated with DingTalk, serving 14M DAU and 800K organizations. Designed an asynchronous messaging architecture supporting 1B messages/day for high-throughput workflows.
- AI Data-Dialogue Platform: Built a high-throughput AI data robot delivering real-time insights across millions of queries/day. Implemented asynchronous task execution with priority queues, Graph DB-managed dependencies, multi-level Map-Reduce, and caching for millions of queries, achieving sub-second latency and improved throughput.
- AI Shop Assistant: Developed an AI shop-assistant robot serving 1M+ merchants, handling 100M+ daily responses and reducing human labor by >90%. Led development of the commercial edition and high-precision billing system.
- Data Service Platform: Architected a scalable data service platform with a data engine and virtualization system, achieving 100K+ peak QPS and 6B+ daily invocations, achieving 99.99% availability with <50ms P99 latency, supporting enterprise-scale AI workloads with low-latency, high-throughput operations.
📚 Blog Posts
- Build Llama3 from Stratch(2025)
- Build Llama2 from Stratch(2025)
- Build, Train and Finetune GPT2 from Stratch(2025)
- Building Enhanced QA with LangChain + LLM (2023)
- Infrastructure Automation Practice: Terraform for Cloud Environment (2022)
- From Distributed Consensus Algorithms to Blockchain Consensus Mechanisms（2019）
- Introduction to the Raft Distributed Consensus Algorithm: Part 1 (2017)
- Introduction to the Raft Distributed Consensus Algorithm: Part 2 (2017)
- Effect of Side-Chain Length on Structural and Dynamic Properties of Ionic Liquids with Hydroxyl Cationic Tails (2014)