Junfan Zhu

Data Structure Algorithms, (GenAI/ML) System Design, Machine Learning, DevOps coding interview practices

Implementations of a Mixture-of-Experts (MoE) architecture designed for research on large language models (LLMs) and scalable neural network designs. One implementation targets a **single-device/NPU environment** while the other is built for multi-device distributed computing. Both versions showcase the core principles.

An efficient and scalable attention module designed to reduce memory usage and improve inference speed in large language models. Designed and implemented the Multi-Head Latent Attention (MLA) module as a drop-in replacement for traditional multi-head attention (MHA) in large language models.

This project demonstrates a decoupled real-time agent architecture that connects LangGraph agents to remote tools served by custom MCP (Modular Command Protocol) servers. The architecture enables a flexible and scalable multi-agent system where each tool can be hosted independently (via SSE or STDIO), offering modularity and cloud-deployable execut

This project implements an advanced Retrieval Augmented Generation (RAG) workflow to enhance question-answering accuracy and reduce LLM hallucinations. It leverages LangGraph to create a stateful, multi-step process that includes document retrieval, relevance grading, and web search fallback.

This project combines two functionalities: a Code Interpreter using LLM Agent Orchestration and Tool Utilization, and a ReAct LangChain Agent example.ReAct agent using LangChain, integrated with a custom tool to calculate the length of a given text. It serves as a foundational example for building more complex, tool-augmented LLM applications.