• 🏫 M.S. in CS @ Sun Yat-sen University (2025-2028)
• 🎯 B.E. in Network Eng. @ Xidian University (2021-2025)
• 🔭 Focus: LLM Fine-tuning (SFT/DPO) & Model Alignment (RLHF)
• 💡 Passion: Clean Arch & DDD

public class Developer {
    private String name = "Chen Yongsai";
    private String[] roles = { "Backend Engineer", "LLM Application Developer" };
    
    public String[] getLocation() {
        return new String[]{"Guangzhou", "Shenzhen"};
    }
}

Microservices Architecture Design & High-Concurrency Community Optimization | Refactoring & Performance Tuning

A comprehensive microservices-based online education platform. Focused on performance bottleneck analysis and system refactoring for the interactive community (comments & likes) and promotion modules.

Highlights:

• 🔄 Interaction Refactoring: Designed Hot/Cold Data Separation: hot data (3 months) in Redis ZSet for ms-level response; cold data in DB with Bloom Filter to block invalid queries (latency 200ms+ → 10ms). Solved high-concurrency write bottlenecks using MQ + Local Write-Behind.
• 🛡️ Concurrency Optimization: Implemented Redisson + Lua double atomic validation to prevent overselling/over-claiming. Used Sequence Space Mapping for massive redemption code generation and BitMap for deduplication, reducing memory by 98% with O(1) checks.
• ⚙️ System Governance: Built Hotspot Dual-Mode Engine (Global Aggregation + Local Caffeine Downgrade). Designed application-layer CDC simulation for ES partial updates to avoid data coverage risks. efficiently handled BigKeys via xxl-job sharded broadcasting and async deletion.

Tech Stack: Spring Boot Spring Cloud MySQL Redis RabbitMQ xxl-job Elasticsearch Caffeine Redisson

AI-Native Dating Recommendation Engine | Solo Full-Stack Project

Because SELECT * FROM users WHERE perfect_match = true never finds true love 😄

An AI-powered dating platform replacing forms with conversational profiling.

Key Innovations:

• 🤖 Agentic Workflow: LangGraph-orchestrated multi-turn dialogue with state persistence
• 🧠 Model Alignment: DeepSeek → SFT/DPO distilled fine-tuning for natural tone
• 🚀 Inference: vLLM + Multi-LoRA (90% cost reduction vs API)
• 🔍 Hybrid Retrieval: MongoDB (Filter) → ElasticSearch (KNN/BM25) → RAG (Psych Ranking)

Tech Stack: Python FastAPI LangGraph vLLM LLaMA Factory Elasticsearch MongoDB

• 🥇 MCM/ICM Finalist (Top 3% globally in Mathematical Contest in Modeling)
• 📜 CET-4/6 Certificates
• 🏅 University Scholarships (4×)

"Code is poetry written in logic. I don't just implement features — I architect solutions, optimize performance, and obsess over clean design patterns."

• 🔬 Deep Diver: Digging into source code & principles
• 📚 Documentation: Official docs > Tutorials
• 🎯 Quality: Strict adherence to standards
• 🤝 Collaboration: Git Flow & Code Review advocate

graph TD
    %% 定义样式类
    classDef ai fill:#e1f5fe,stroke:#01579b,stroke-width:2px,color:#01579b;
    classDef be fill:#fff3e0,stroke:#e65100,stroke-width:2px,color:#e65100;
    classDef goal fill:#e8f5e9,stroke:#1b5e20,stroke-width:4px,color:#1b5e20;

    subgraph LLM_Path [🤖 AI & LLM Specialization]
        direction TB
        A[LLM Fundamentals]:::ai --> B[Fine-tuning SFT/DPO]:::ai
        B --> C[Model Alignment RLHF]:::ai
        C --> D[RAG & Vector DB]:::ai
        D --> E[Agentic Workflows]:::ai
    end

    subgraph Backend_Path [🛠️ Backend & Architecture]
        direction TB
        F[Business Logic & CRUD]:::be --> G[Microservices & RPC]:::be
        G --> H[DDD & Clean Arch]:::be
        H --> I[High Concurrency/Performance]:::be
        I --> J[Distributed System Design]:::be
    end

    %% 两条路最终汇聚的目标
    E --> K[Build AI-Native Applications]:::goal
    J --> K