From AI Coding to Full‑Stack AI Apps: Master Claude, Codex, Agents, and Skills

What Is AIGuide

I am the author of JavaGuide and have compiled the past few months of work on AI application development, AI coding practice, and AI interview preparation into an open‑source guide called AIGuide . All content is freely available and can be cloned for local reading.

The guide targets developers who want a systematic path to AI application development without first becoming a research‑oriented algorithm engineer. It focuses on the parts that matter in real projects: LLM fundamentals, RAG, agents, prompts, evaluation, system design, Claude Code, Codex, and related tooling.

How to Read the Guide

LLM Basics : Understand tokens, context windows, sampling parameters, API usage, structured output, and evaluation.

RAG : Learn document chunking, vector retrieval, update pipelines, and evaluation for enterprise knowledge‑base Q&A.

AI Agent : Dive into tool calling, memory, MCP, skills, and workflow/graph/loop designs.

AI System Design : Move a demo into production, handling gateways, rate‑limiting, fallback, cost, observability, security, and gradual rollout.

Why Create AIGuide

JavaGuide has been maintained since May 2018 and is well known among Java developers. Recent questions about the relevance of knowledge‑base projects in the age of AI prompted a reflection: AI answers are fast but not always accurate, often mixing outdated or non‑recommended solutions, and they may miss real‑world constraints such as deployment feasibility.

For example, using RAG effectively requires decisions about document chunking, vector store selection (HNSW, IVFFLAT, pgvector, Milvus, Elasticsearch), handling PDF parsing errors, update deduplication, hybrid search, query rewriting, and reranking—all of which impact code, data, and metrics.

AIGuide aims to collect such detailed, engineering‑level knowledge—things that AI models often get wrong—into a developer‑friendly resource.

Contents of AIGuide

LLM Fundamentals

Common pitfalls such as blindly tweaking temperature, context length, or max tokens are explained as engineering parameters. Sample questions include token consumption differences between Chinese and English, why structured output can still fail with temperature 0, and how to add timeout, retry, cancellation, and idempotency logic to LLM API calls.

AI Agent

The guide starts from basic concepts (agent loop, plan‑and‑execute, A2A, agentic workflows, tool registration) and then expands to memory placement, prompt vs. context engineering, skills, MCP, function calling, and how to contain agent nondeterminism with workflow/graph/loop structures. Prompt‑injection attacks are also covered, with mitigation techniques such as context isolation, permission gating, human confirmation, and output filtering.

RAG

RAG is presented beyond the three‑step naïve view (document chunk → vector → retrieval). It discusses PDF parsing order, chunk size trade‑offs, vector database choices, document update handling (deduplication, gray‑release, rollback), hybrid search, query rewrite, and rerank placement. The guide separates document processing, vector DB selection, knowledge‑base updates, GraphRAG, retrieval optimization, and RAG evaluation.

AI System Design

Production challenges such as model timeouts, token cost spikes, prompt regressions, missing permission filters in RAG results, and agent tool failures are examined. Topics include multi‑model routing, fallback, rate‑limiting, billing, chaining of prompt, RAG, memory, tools, evaluation, security, cost monitoring, observability, audit, and gray‑release strategies. Real‑time voice agents (ASR, TTS, VAD, interruption handling, low‑latency optimization) are also discussed.

AI Coding Practice

Practical scenarios demonstrate how AI coding tools can be used in daily work: IDEA + Qoder for interface optimization, Trae + MiniMax for Redis fault diagnosis, Claude Code with GLM‑5.1 for JVM diagnostics, DeepSeek V4 + Claude Code for code audit and database migration, MiniMax M3 + Claude Code for Redis troubleshooting, Claude Desktop with CC Switch for third‑party model integration, and JetBrains plugins for Claude Code/Codex GUI enhancements.

The focus is on whether the AI can understand project context, modify code correctly, run verification, and help locate errors.

AI Coding Tips

Tool selection (Claude Code, Codex, Cursor, Trae, Qoder) and workflow breakdown are covered. Examples include configuring Claude Code with sub‑agents and multi‑instance collaboration ( CLAUDE.md), using commands such as simplify, code-review, loop, batch, run, verify for different tasks, applying TDD, code review, and web‑automation skills to improve AI coding experience, and writing clear specifications for AI‑driven coding.

The overall goal is to rely less on intuition and more on context, rules, verification, and retrospection.

Learning Paths and Interview Questions

The README provides separate sections for learning routes (2026‑latest roadmaps for Java/Go developers, backend‑to‑AI‑agent transition, full‑stack backend learning) and interview question collections (AI application development, LLM basics, agents, RAG, system design).

These resources are intended as study guides rather than rote memorization; learners should follow the roadmap, use the interview questions to identify gaps, and be able to discuss the material in the context of real projects.

Why AIGuide Belongs to the JavaGuide Ecosystem

JavaGuide serves Java and backend developers; AIGuide serves anyone wanting a systematic AI application development path. The standards and quality expectations are shared, and the audience now includes backend, frontend, testing, and product‑technical roles.

Because AI tooling and frameworks evolve quickly, the guide will continue to be updated, but the current version provides a solid backbone to avoid scattering across disparate resources.

Conclusion

AIGuide does not merely stack AI concepts; it is an engineering‑focused learning material that first explains LLM operation, then covers RAG, agents, prompts, evaluation, and system design, and finally integrates AI coding into everyday development workflows.