Focused on deep research and development of large-model applications. Authors of "RAG Application Development and Optimization Based on Large Models" and "MCP Principles Unveiled and Development Guide". Primarily B2B, with B2C as a supplement.
This article explains what overfitting is, introduces dropout as a regularization technique, describes how dropout randomly deactivates neurons during training and rescales outputs during inference, discusses its limitations, and outlines why large language models may use alternative strategies.
This article explains how to construct a Retrieval‑Augmented Generation (RAG) pipeline for multi‑page PPT documents by converting slides to images, extracting content with a vision model, indexing with LlamaIndex and Chroma, and exposing the functionality through an MCP Server with tools for adding, querying, and managing PPTs.
This guide explains Google’s open‑source MCP Toolbox for Databases, covering its core concepts, installation, configuration, two usage modes (native SDK and MCP), example LangGraph agent integration, security features, observability, and practical code snippets for building reliable LLM‑driven database tools.
This article explains how to extend a single‑process Human‑in‑the‑Loop (HITL) agent to a distributed, multi‑user API service using FastAPI, detailing session management, interrupt handling, client and server fault‑recovery strategies, and providing concrete code snippets and architectural diagrams.
This article explains the mathematical foundation of layer normalization, why it is needed for deep neural networks like Transformers, how scaling (γ) and bias (β) parameters restore important signal variations, and practical placement tips for stable training.
This article explains why Human-in-the-Loop (HITL) is essential for enterprise LLM agents, outlines common HITL patterns, and shows how LangGraph’s interrupt, resume, and checkpoint mechanisms can be used to build reliable, auditable workflows with tool‑call control and practical code examples.
This guide provides a comprehensive walkthrough of the new streamable HTTP transport mode introduced in MCP SDK v1.9.0, covering server and client configuration, core parameters, session‑id handling, sample implementations, multi‑instance deployment, and practical testing results.
This article explains why residual connections are essential in deep neural networks, describing the problems of network degradation and gradient vanishing, how shortcut paths add the input to the layer output, the requirement of matching dimensions, and the resulting stability for training large language models.
This article examines five major LLM‑driven AI agent planning and execution patterns—ReAct, Plan‑and‑Execute, Static Workflow, Static Workflow with local intelligence, and modular hierarchical planning—detailing their mechanisms, code examples, strengths, weaknesses, suitable scenarios, and optimization techniques.
The updated MCP protocol (2025‑03‑26) introduces JSON‑RPC batch processing and enhanced tool annotations, enabling AI agents to send multiple requests in a single call, improve performance, and provide richer metadata for safer, more transparent tool usage.
