Open to opportunities

wenjun li

@wenjunli

Message

I build GenAI multimodal and RAG systems that improve recommendation and moderation outcomes.

Zimbabwe

Message

What I'm looking for

I’m looking for a team where I can architect GenAI/LLM and multimodal RAG systems end-to-end, with clear metrics (CTR, accuracy, cost) and strong cross-functional collaboration—shipping production impact from model design to optimized inference.

I’m a Senior GenAI-focused software architect who turns multimodal model advances into production-ready pipelines. I lead LLM/GenAI adoption for content understanding, unifying outputs across recommendation, profiling, moderation, and operations teams.

Across RAG, vector search, and agentic workflows, I design end-to-end architectures from preprocessing and training to inference serving. I’ve delivered measurable gains such as improving accuracy from 79% to 91%, cutting review costs by over ¥20M/year, and reducing iteration cycles from weeks to hours without retraining.

Experience

Work history, roles, and key accomplishments

Current

Senior Algorithm Engineer

Current

NetEase

Dec 2021 - Present (4 years 6 months)

Led a company-wide multimodal content understanding platform and GenAI/AIGC applications, evolving discriminative models to generative large models (Qwen-VL/InternVL) for image-text/video classification, tagging, and quality assessment. Improved accuracy from 79% to 91% and reduced review costs by over ¥20M/year; also built an agentic LangChain/LangGraph pipeline that cut iteration cycles from wee

GenAI RAG Lang Chain LangGraph VLLM Contrastive Learning

Senior Algorithm Engineer

Kuaishou

Jul 2021 - Sep 2021 (2 months)

Implemented SBERT-based semantic similarity for trending Kuaishou content and used Faiss to cluster embedded text for downstream event timeline and event relationship generation.

Embeddings Faiss Clustering NLP Event Modeling Vector Search

Algorithm Engineer

Sina Weibo

Feb 2016 - Jun 2021 (5 years 4 months)

Built Weibo’s vertical account system including user profiling, hierarchy construction, and tagging, and supported high-quality user acquisition across platforms. Developed ranking computations for top influencers to drive vertical operations.