Engineering Notes

Writing about what I build

13 posts on agentic AI, RAG, embeddings, MCP, and production patterns.

agentic-airagmcp

The Agentic AI Stack — Every Layer, Mapped

Embeddings, vector stores, orchestration, memory, guardrails, observability. A full map of the agentic AI landscape and where each piece fits.

agentic-ailearningcareer

6 Months of Building AI Agents — What I Learned

16 years of frontend engineering meets agentic AI. What actually matters, what surprised me, and where I'm going next.

tool-useagentic-aianthropic

Tool Use — The LLM Decides, Your Code Executes

A deep dive into how tool calling actually works — the request shape, response shape, and the critical insight most tutorials miss.

productionagentic-aibullmq

Production AI Agents — Queues, Rate Limits, and Cost Control

What separates a demo from production. The patterns that make AI agents reliable, affordable, and scalable.

agentic-aiworkflowllm

Workflow vs LLM Agent — When to Use Each

Not every problem needs an AI agent. Here's the honest trade-off and the production sweet spot nobody talks about.

Tokens — What They Are and Why They Drive Every LLM Decision

Tokens are not words. Not characters. Understanding them changes how you design, optimize, and price AI systems.

ragchunkingembeddings

Chunking — How to Split Documents for RAG

The strategy you use to split documents matters more than the embedding model you choose. Here's what I learned.

embeddingsragvectors

Embeddings — How Text Becomes Numbers That Capture Meaning

What embeddings are, why they matter for RAG, and how 'can't reach the work network' matches VPN documentation.

mcpagentic-aianthropic

MCP — The USB-C Standard for AI Tools

Model Context Protocol is how AI agents connect to the world. Anthropic created it, OpenAI and Google adopted it. Here's what it means for engineers.

agentic-aiai-agentllm

Agentic AI vs AI Agent — What's the Real Difference?

Most people use these terms interchangeably. They're not the same. Here's the distinction that matters in production.

agentic-airagtool-use

Building an IT Support Agent — RAG + Parallel Tool Calls

How I built an AI agent that diagnoses and fixes VPN and tool access issues using RAG and parallel tool execution.

RAG — You Don't Always Need a Vector Database

How to build Retrieval Augmented Generation with just a JSON array. Start simple, add complexity only when you need it.

agentic-aitool-usenodejs

The Agentic Loop — How AI Agents Actually Work

The while loop that powers every serious AI agent. From calendar scheduling to GitHub Copilot — it's all the same pattern.