Series
Series
Scattered deep-dives organized into structured learning paths.
5 series · 16 posts
01 → 02 → 03 → 04 → 05 →
Harness Engineering
Posts in the Harness Engineering series
- Harness Engineering: The Model Isn't Dumb, It Just Lacks Human Guidance
- Harness Engineering (2): Five Engineering Answers from OpenAI's Million-Line Experiment
- Harness Engineering (3): Industry Consensus, Four Pillars, and a Three-Phase Rollout
AI Agent 實戰
A curated reading path (not a serialized course): deep-dives on AI agent engineering — from harness and the agent loop to loop design and cost control.
- Unrolling the Codex Agent Loop: What Actually Happens Behind a Single Conversation Turn
- Loop Engineering: Designing Systems That Prompt Agents for You
- How to Design Safety Layers for an AI Agent: From Keyword Detection to Long-Term Behavioral Monitoring
RAG 系統架構
A curated reading path (not a serialized course): rethinking RAG, vector vs reasoning retrieval, and the context-rot / compression problem.
- Beyond RAG: A Deep Dive into Andrej Karpathy's LLM Wiki Pattern
- RAG's Five Stages: From Pipeline to Reasoning Retrieval, and the Naive RAG on My Own Site
- Building a Real RAG: 5 Infra Lessons from InfiniFlow's 2024 Year-in-Review
- +2 more →
Claude Code 自動化指南
A curated reading path (not a serialized course): from getting Claude Code running and the dialogue-as-doc workflow to MCP integration and skills/tools loading.
- Dialogue as Documentation: Turning a Debug Session into an Article with Claude Code
- MCP in Claude Code: How Model Context Protocol Connects AI to Your Tool Ecosystem
- Is Claude Code's On-Demand Loading of Skills/Tools a Form of RAG? Unpacking Agentic Retrieval
Kafka 為什麼這麼快
A two-part deep dive into what makes Kafka fast — from sequential I/O and zero-copy to partitions and consumer groups.
- Why Is Kafka Fast? Sequential I/O and Zero-Copy Explained
- Why Is Kafka Fast? Part 2: Partitions, Replication, and Consumer Groups