Skip to content
View Zsyyxrs's full-sized avatar
🎯
Focusing
🎯
Focusing
  • shanghai

Block or report Zsyyxrs

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Zsyyxrs/README.md

Hi, I'm Shangyi Zhu 👋

AI Algorithm Engineer based in Shanghai, focused on LLM applications & Agent systems.

7 years of engineering experience — 5 years building large-scale data infrastructure at NIO (autonomous driving data) and SAIC-GM (intelligent connected vehicles), and 2+ years transitioning into LLM application development. Currently building a multi-agent Q&A platform for the banking domain.

I work at the intersection of data engineering and LLMs — turning messy enterprise data into reliable AI products.


🔭 What I'm working on

  • Bank Intelligent Q&A Platform — Multi-agent architecture with NL2SQL, RAG, and tool-calling routing. Built on LangGraph with BGE-M3 hybrid retrieval and Qwen-series models.
  • Medical RAG System — Tri-modal hybrid retrieval (dense + sparse + ColBERT) over Milvus/HNSW, with LangGraph orchestration and RAGAS evaluation. ChatGLM3-6B fine-tuned with LoRA/QLoRA.
  • Exploring — Multimodal agents, agent evaluation frameworks, and production deployment patterns (vLLM, quantization).

🛠️ Tech Stack

LLM & Agents LangGraph · LangChain · RAG · NL2SQL · vLLM · LoRA/QLoRA · RAGAS

Models & Frameworks Qwen · ChatGLM · BGE-M3 · PyTorch · Transformers · PEFT

Data & Infra Milvus · Doris · ClickHouse · Hive · Iceberg · Spark · Flink

Languages & Tools Python · SQL · Java · Docker · Git · Linux


📌 Selected Projects

Project Stack Highlights
Legal RAG System FastAPI, ChromaDB, bge-zh, Qwen3 Statute-aware article-level chunking, hybrid BM25+dense retrieval, query rewriting, cited answers with confidence scoring
Positive Chinese Chatbot PyTorch, ChatGLM3-6B, LoRA, Gradio LoRA fine-tuning on Douban "夸夸" corpus for encouraging replies, Trie-based dirty-word filter with variant/homophone detection, multi-strategy decoding with BLEU/ROUGE/diversity evaluation
Chinese Sentiment Classifier PyTorch, BERT-base-Chinese, Transformers, ChnSentiCorp Selective last-N layer unfreezing on BERT with weighted sampling and jieba synonym augmentation, AMP + warmup-linear LR + early stopping, MC-dropout uncertainty and attention-based explainability, 94.5% Acc / AUC 0.983

📫 Get in touch

Open to opportunities in LLM / Agent engineering — particularly roles involving data agents, RAG systems, or multimodal applications.

Feel free to reach out via GitHub.

Pinned Loading

  1. zh-sentibert zh-sentibert Public

    基于 BERT-base-Chinese 微调的中文情感分类器:在 ChnSentiCorp 上采用末层选择性解冻、加权采样 + jieba 同义词增强、混合精度训练与 warmup-linear 学习率调度;内置 MC-dropout 不确定性估计与注意力可解释性,测试集 Accuracy 94.5% / AUC 0.983。

    Python 1

  2. chatbot chatbot Public

    基于 PyTorch 与 LoRA 微调 ChatGLM3-6B 的中文聊天机器人,专注生成积极正面的回复,内置敏感词过滤、多策略解码与 BLEU/ROUGE 评估,支持 CLI、Gradio Web 与批量推理三种部署方式。

    Python

  3. legal-rag-system legal-rag-system Public

    基于检索增强生成(RAG)的中文法律智能问答系统:自然语言提问,按"第 X 条"级别精准检索,答案附带法条引用与置信度。支持语义+BM25 混合检索、查询改写、领域过滤,提供 Web 界面与 REST API。

    Python