https://sinimite.work/en/Recent content on SiBloghttps://sinimite.work/images/og-default.svg?v=20260525-210321https://sinimite.work/images/og-default.svg?v=20260525-210321Hugo -- 0.156.0en-USFri, 17 Jul 2026 15:40:44 +0900https://sinimite.work/en/posts/rag-retrieval-engineering-guide/Fri, 17 Jul 2026 15:40:44 +0900https://sinimite.work/en/posts/rag-retrieval-engineering-guide/A technical report on designing and implementing the RAG retrieval layer, covering BM25, embeddings, ANN, RRF, reranking, query understanding, access control, evaluation, and deployment practices.https://sinimite.work/en/posts/rag-observability-evaluation-design/Fri, 17 Jul 2026 15:06:18 +0900https://sinimite.work/en/posts/rag-observability-evaluation-design/A systematic guide to RAG observability and evaluation, including staged traces, runtime metrics, evaluation datasets, quality scoring, experiment comparison, privacy boundaries, and graceful degradation.https://sinimite.work/en/posts/opentelemetry-ai-agent-observability-guide/Thu, 16 Jul 2026 21:08:52 +0900https://sinimite.work/en/posts/opentelemetry-ai-agent-observability-guide/A beginner&#39;s guide to OpenTelemetry for AI application developers, covering traces, metrics, logs, profiles, context propagation, the Collector, GenAI semantic conventions, and production practices.https://sinimite.work/en/posts/rag-chunking-strategies-evaluation/Wed, 15 Jul 2026 18:05:16 +0900https://sinimite.work/en/posts/rag-chunking-strategies-evaluation/A technical reference covering RAG chunking strategies, parameter selection, context preservation, and evaluation workflows.https://sinimite.work/en/posts/rag-rrf-retrieval-fusion/Wed, 15 Jul 2026 17:37:44 +0900https://sinimite.work/en/posts/rag-rrf-retrieval-fusion/A technical reference on the principles of Reciprocal Rank Fusion, its place in the retrieval pipeline, fusion choices, and production evaluation.https://sinimite.work/en/posts/rag-dense-sparse-hybrid-search/Wed, 15 Jul 2026 17:25:30 +0900https://sinimite.work/en/posts/rag-dense-sparse-hybrid-search/A technical reference on Dense + Sparse Hybrid Search, including its principles, fusion methods, suitable use cases, and evaluation approach.https://sinimite.work/en/posts/enterprise-rag-system-building-guide/Wed, 15 Jul 2026 16:35:23 +0900https://sinimite.work/en/posts/enterprise-rag-system-building-guide/An introduction to the architecture of enterprise RAG data ingestion, hybrid retrieval, access control, generation, evaluation, and observability.https://sinimite.work/en/posts/deep-research-aiq-skill-agent-harness-mcp/Mon, 25 May 2026 11:00:00 +0900https://sinimite.work/en/posts/deep-research-aiq-skill-agent-harness-mcp/<p>On May 20, 2026, NVIDIA published a technical article explaining how to package AI-Q&rsquo;s deep-research capability as a “specialized skill” callable by agent harnesses such as Claude Code, Codex, and OpenCode. The key point is not merely that another AI tool exists, but that the design proposes a clearer separation for enterprise Agents: a general-purpose agent harness manages conversation, tool orchestration, code execution, and user interaction, while a specialized research backend handles multisource retrieval, planning, synthesis, citations, evaluation, and enterprise data governance. (<a href="https://developer.nvidia.com/blog/add-a-specialized-deep-research-skill-to-agent-harnesses/" title="Add a Specialized Deep Research Skill to Agent Harnesses | NVIDIA Technical Blog">NVIDIA Developer</a>)</p> <h2 id="1-background-why-shouldnt-a-general-purpose-agent-perform-deep-research-directly">1. Background: Why Shouldn&rsquo;t a General-Purpose Agent Perform Deep Research Directly?</h2> <p>Harnesses such as Claude Code, Codex, and LangChain Deep Agents are effective interaction entry points for developers. They maintain conversations, invoke tools, execute code, and turn user intent into action chains. But when the task becomes “generate a cited research report from multiple enterprise documents, internal databases, external materials, and regulated data sources,” the complexity quickly grows from “call several tools” into “build a complete research pipeline.” NVIDIA explicitly notes that enterprise teams must address data access, authentication, query routing, prompt tuning, output evaluation, and citation fidelity, and that these concerns should not be reimplemented in every harness. (<a href="https://developer.nvidia.com/blog/add-a-specialized-deep-research-skill-to-agent-harnesses/" title="Add a Specialized Deep Research Skill to Agent Harnesses | NVIDIA Technical Blog">NVIDIA Developer</a>)</p>https://sinimite.work/en/posts/points-of-the-podcast-language-agents-from-reasoning-to-acting/Sun, 24 May 2026 16:05:10 +0900https://sinimite.work/en/posts/points-of-the-podcast-language-agents-from-reasoning-to-acting/Reflections on the podcast Language Agents: From Reasoning to Acting. Starting from ReAct, Reflexion, Tree of Thoughts, memory, benchmarks, and ACI, the article explains why an LLM Agent is an engineered system composed of a model, tools, memory, environment, evaluation, and UX.https://sinimite.work/en/posts/ai-engineer-skill-value-map-2026/Sun, 03 May 2026 20:09:15 +0900https://sinimite.work/en/posts/ai-engineer-skill-value-map-2026/A skill value map for engineers moving into AI application engineering, explaining how the value of basic RAG, prompt engineering, and framework APIs is changing, and why evaluation, governance, and agentic workflows deserve greater investment.https://sinimite.work/en/posts/work-schedule-optimization-manual/Sun, 03 May 2026 00:00:00 +0900https://sinimite.work/en/posts/work-schedule-optimization-manual/This article covers schedule adjustment, deep work, the afternoon slump, a bedtime shutdown routine, and a 14-day experiment. Its central goal is to align tasks, energy, and time.https://sinimite.work/en/posts/ai-agent-architecture-design-guide/Sat, 02 May 2026 00:00:00 +0000https://sinimite.work/en/posts/ai-agent-architecture-design-guide/A practical guide to AI coding agents explaining how specs, fat skills, a thin harness, deterministic tooling, and a learning loop create reliable, continuously improving AI agent systems.https://sinimite.work/en/posts/llm-from-hidden-state-to-token-output/Sat, 02 May 2026 00:00:00 +0000https://sinimite.work/en/posts/llm-from-hidden-state-to-token-output/Why does Chain of Thought work? Why are longer prompts not always better? Why do models hallucinate? These apparently unrelated engineering phenomena are different expressions of the same core mechanism. This article uses one master key to open five doors in LLM engineering.https://sinimite.work/en/posts/ai-coding-agent-software-spec-best-practices/Thu, 09 Apr 2026 20:03:59 +0900https://sinimite.work/en/posts/ai-coding-agent-software-spec-best-practices/A guide to writing software specifications in the age of AI coding agents, covering the definition of a Software Spec, its six core areas, a four-stage workflow, best practices, and a reusable template.https://sinimite.work/en/posts/llm-training-for-ai-engineers/Sat, 04 Apr 2026 12:00:00 +0900https://sinimite.work/en/posts/llm-training-for-ai-engineers/A panoramic guide to LLM training for AI application engineers, covering pretraining, post-training, distillation, reward design, agent training, and harness engineering, so you can understand where model capabilities come from and how training decisions affect real-world deployment.https://sinimite.work/en/posts/agent-native-documentation-engineering/Sun, 29 Mar 2026 16:55:53 +0900https://sinimite.work/en/posts/agent-native-documentation-engineering/A guide to documentation architecture for AI coding agent projects, covering the layered responsibilities of AGENTS.md, PRD, Architecture, Spec, and Plan documents, plus the organizational principles and engineering value of agent-native documentation.https://sinimite.work/en/posts/ai-coding-agent-documentation-best-practices/Sat, 28 Mar 2026 00:00:00 +0000https://sinimite.work/en/posts/ai-coding-agent-documentation-best-practices/Drawing on OpenAI&#39;s Harness Engineering, Anthropic&#39;s research on long-running agents, the AGENTS.md standard, and experience from multiple production teams, this article presents best practices for designing project documentation for AI coding agents.https://sinimite.work/en/posts/llm-agentic-evolution/Sat, 28 Mar 2026 00:00:00 +0000https://sinimite.work/en/posts/llm-agentic-evolution/LLMs are evolving from passive question-answering tools into agents that pursue goals autonomously. This is not a linear increase in model capability but a fundamental change in the usage paradigm. This article examines the nature of that transition, the evolution of the engineering stack, and what it means for engineers.https://sinimite.work/en/posts/ai-agent-standards-full-landscape/Sat, 28 Mar 2026 00:00:00 +0000https://sinimite.work/en/posts/ai-agent-standards-full-landscape/At the beginning of 2026, standardization in the AI Agent field is advancing at astonishing speed. Four core standards—MCP, A2A, AGENTS.md, and SKILL.md—have each found their place, and AAIF has begun operating as a governance organization. Using analogies to the Java ecosystem, this article helps backend engineers develop a panoramic view of the standards landscape.https://sinimite.work/en/posts/rag-system-complete-guide/Fri, 27 Mar 2026 15:06:20 +0900https://sinimite.work/en/posts/rag-system-complete-guide/A panoramic RAG guide covering ingestion, chunking, retrieval, reranking, generation, evaluation, and production operations, with an implementation path from zero to launch.https://sinimite.work/en/posts/entropy-in-llm/Fri, 27 Mar 2026 00:00:00 +0000https://sinimite.work/en/posts/entropy-in-llm/Entropy runs through the entire LLM lifecycle: training uses cross-entropy as the loss, inference uses temperature to control output entropy, evaluation uses perplexity, and products can use entropy signals to detect hallucinations. This article maps the complete picture from an AI engineer&#39;s perspective.https://sinimite.work/en/posts/spec-driven-development-guide/Thu, 19 Mar 2026 00:01:55 +0900https://sinimite.work/en/posts/spec-driven-development-guide/A systematic guide to the definition, workflow, tooling ecosystem, and appropriate boundaries of Spec-Driven Development, explaining why it is replacing Prompt-and-Pray AI development.https://sinimite.work/en/posts/rag-chunking-best-practices/Wed, 18 Mar 2026 20:40:55 +0900https://sinimite.work/en/posts/rag-chunking-best-practices/A practical RAG chunking guide for AI application engineers, covering default parameters, mainstream splitting strategies, evaluation workflows, and production practices.https://sinimite.work/en/posts/agent-model-harness/Wed, 18 Mar 2026 00:17:00 +0900https://sinimite.work/en/posts/agent-model-harness/Starting from the formula Agent = Model + Harness, this article reframes the core work of an AI application engineer as harness engineering.https://sinimite.work/en/posts/python-concurrency-parallelism-complete-guide/Sun, 15 Mar 2026 10:00:00 +0900https://sinimite.work/en/posts/python-concurrency-parallelism-complete-guide/A practical guide to concurrency and parallelism in Python, covering the GIL, multithreading, multiprocessing, asyncio, thread safety, and the evolution of Python&#39;s concurrency ecosystem in Python 3.14.https://sinimite.work/en/posts/python-async-iteration-complete-guide/Sun, 15 Mar 2026 09:00:00 +0900https://sinimite.work/en/posts/python-async-iteration-complete-guide/A practical review guide to Python&#39;s async and iteration systems, covering iterable, iterator, generator, await, async for, and thread-pool bridging.https://sinimite.work/en/posts/prompt-engineering-best-practices-guide/Wed, 11 Mar 2026 11:00:00 +0900https://sinimite.work/en/posts/prompt-engineering-best-practices-guide/A systematic guide to production-grade Prompt Engineering and Context Engineering practices for 2025–2026, covering Claude, GPT, long context, Prompt Chaining, Tool Use, and evaluation workflows.https://sinimite.work/en/posts/anthropic-contextual-retrieval-reading-notes/Wed, 11 Mar 2026 09:00:00 +0900https://sinimite.work/en/posts/anthropic-contextual-retrieval-reading-notes/Reading notes on Anthropic&#39;s Contextual Retrieval, covering semantic loss after chunking, experimental benchmarks, ground-truth evaluation, and production RAG practices.https://sinimite.work/en/posts/contextual-retrieval-rag-reading-list/Mon, 09 Mar 2026 08:30:00 +0900https://sinimite.work/en/posts/contextual-retrieval-rag-reading-list/A Contextual Retrieval and RAG reading list for AI application engineers, covering official Anthropic resources, open-source implementations, evaluation methods, and industry trends.https://sinimite.work/en/posts/apple-m5-vs-m4-practical-comparison-ai-engineers/Fri, 06 Mar 2026 08:30:00 +0900https://sinimite.work/en/posts/apple-m5-vs-m4-practical-comparison-ai-engineers/An engineering comparison of M5 and M4 covering single- and multi-core performance, cache, unified memory, GPU AI acceleration, and the local-inference experience.https://sinimite.work/en/posts/llm-api-kv-cache/Thu, 05 Mar 2026 11:09:00 +0900https://sinimite.work/en/posts/llm-api-kv-cache/KV Cache is a key concept connecting Transformer theory with LLM engineering and deployment. Understanding it completes the path from how a model computes to how it runs.https://sinimite.work/en/about/Thu, 05 Mar 2026 09:00:00 +0900https://sinimite.work/en/about/A brief introduction to this site and its author.https://sinimite.work/en/posts/llm-chain-of-thought-cot/Wed, 04 Mar 2026 21:51:00 +0900https://sinimite.work/en/posts/llm-chain-of-thought-cot/Understand what LLM Chain-of-Thought (CoT) is and how prompt engineering can elicit Chain-of-Thought (CoT) from an LLM.https://sinimite.work/en/posts/prompt-engineering-from-concept-to-implementation/Tue, 03 Mar 2026 19:50:00 +0900https://sinimite.work/en/posts/prompt-engineering-from-concept-to-implementation/Prompt engineering? It may not be as simple as you think.https://sinimite.work/en/posts/understanding-prompt-injection/Tue, 03 Mar 2026 19:50:00 +0900https://sinimite.work/en/posts/understanding-prompt-injection/Understand LLM prompt injection and several fundamental defensive measures.https://sinimite.work/en/posts/understanding-transformer-intuition/Tue, 24 Feb 2026 12:00:00 +0900https://sinimite.work/en/posts/understanding-transformer-intuition/My first formal blog post.