Agentic AI Reasoning Model System Design

Companion project for Agentic AI Reasoning Model System Design (AI Engineering Insider). A runnable reasoning system covering decomposition, planning, search, tools, GraphRAG, verification, adaptive test-time compute, and a multi-agent supervisor, served by FastAPI + LangGraph behind a Next.js UI, sized to run on a laptop with Ollama.

Every measurement in the book can be reproduced here. That is the point of the project: the labs exist so you can watch the behaviour rather than trust the prose.

The project is structured into hands-on labs:

↳ Reasoning Fundamentals

↳ Decomposition

↳ Planning

↳ Search

↳ Tools & Actions

↳ Retrieval + GraphRAG

↳ Verification

↳ Test-Time Compute

↳ Multi-Agent

↳ Production