Start here ยท Interactive map of the series
From Transformer to Production Agentic AI: The Complete Stack
Twenty components, one system. Press play to follow a real request through every layer, or watch a model being built from raw data. Click any component to see what it does, a result from this series' experiments, and its deep dive.
build ยท request ยท feedback
Tip: click any component, or press play.
โ Build the model offline
โก Give it context and capabilities per request
โข Run it in production platform
โฃ Keep it safe and improving always on
Every Deep Dive in the Series
๐Pretraining
๐คTokenization & BPE
๐ง How a Transformer Works
๐๏ธFine-tuning
๐๏ธPost-training
๐ฅ๏ธScaling & Distributed Training
โกHow Inference Works
๐งฉContext Engineering & Memory
๐RAG, End to End
๐ ๏ธTool Calling
๐ ๏ธMCP
๐๏ธMemory & Planning Patterns
๐ฅMulti-Agent Systems
๐ฅAgentic Systems
๐คProduction System Design
๐๏ธProduction Agent Architecture
๐Evaluating Agents
About the numbers: each "from the series" figure is a real result from the runnable experiments or verified sources in the linked article; the article explains the setup and its limits. The request journey shows the common architecture of production agents, not the internals of any specific product.