How shayanthn/ai-agent-architectures works
This repository is an educational collection of runnable Jupyter demonstrations, not a single deployed agent service. The principal workflow is for a developer to choose and run an architecture notebook to explore its agent pattern; the notebooks cover foundational, collaborative, memory/reasoning, and safety or self-improvement designs. The README identifies LangGraph as the core orchestrator and describes LLM-as-a-Judge evaluation as a capability of most examples. Notebook internals were not sampled, so the graph does not assert specific calls between their components or identify which notebooks implement evaluation.
Main entry points: 01_reflection.ipynb Reflection · 02_tool_use.ipynb Tool Use · 03_ReAct.ipynb ReAct · 04_planning.ipynb Planning
Components 19
Foundational Patterns
- Reflection
01_reflection.ipynb - Tool Use
02_tool_use.ipynb - ReAct
03_ReAct.ipynb - Planning
04_planning.ipynb
Agent Collaboration
- Multi-Agent
05_multi_agent.ipynb - Blackboard
07_blackboard.ipynb - Meta-Controller
11_meta_controller.ipynb - Ensemble
13_ensemble.ipynb
Memory And Reasoning
- Episodic Semantic Memory
08_episodic_with_semantic.ipynb - Tree Of Thoughts
09_tree_of_thoughts.ipynb - World-Model Graph
12_graph.ipynb - Cellular Automata
16_cellular_automata.ipynb
Safety And Improvement
- Plan Execute Verify
06_PEV.ipynb - Mental Loop
10_mental_loop.ipynb - Dry-Run Harness
14_dry_run.ipynb - RLHF
15_RLHF.ipynb - Reflexive Metacognition
17_reflexive_metacognitive.ipynb
Other components
- Developer
- LangGraph
Connections 34
- Developer → Reflection: runs
- Developer → Tool Use: runs
- Developer → ReAct: runs
- Developer → Planning: runs
- Developer → Multi-Agent: runs
- Developer → Plan Execute Verify: runs
- Developer → Blackboard: runs
- Developer → Episodic Semantic Memory: runs
- Developer → Tree Of Thoughts: runs
- Developer → Mental Loop: runs
- Developer → Meta-Controller: runs
- Developer → World-Model Graph: runs
- Developer → Ensemble: runs
- Developer → Dry-Run Harness: runs
- Developer → RLHF: runs
- Developer → Cellular Automata: runs
- Developer → Reflexive Metacognition: runs
- Reflection → LangGraph: orchestrates with
- Tool Use → LangGraph: orchestrates with
- ReAct → LangGraph: orchestrates with
- Planning → LangGraph: orchestrates with
- Multi-Agent → LangGraph: orchestrates with
- Plan Execute Verify → LangGraph: orchestrates with
- Blackboard → LangGraph: orchestrates with
- Episodic Semantic Memory → LangGraph: orchestrates with
- Tree Of Thoughts → LangGraph: orchestrates with
- Mental Loop → LangGraph: orchestrates with
- Meta-Controller → LangGraph: orchestrates with
- World-Model Graph → LangGraph: orchestrates with
- Ensemble → LangGraph: orchestrates with
- Dry-Run Harness → LangGraph: orchestrates with
- RLHF → LangGraph: orchestrates with
- Cellular Automata → LangGraph: orchestrates with
- Reflexive Metacognition → LangGraph: orchestrates with