1 of 10

Building your own agentic system

2 of 10

Why building your own agentic system?

�

  • You control the flow, tools, and depth

​

  • You control the LLMs to use (open or close source)�
  • You need to run the system many times and need a reproducible pipeline�
  • Control what is in the context of each agent�
  • We have control over applications to science�
  • Is fun!

�

3 of 10

Task

4 of 10

  • AI agents / Non-AI agents

​

  • Tools

​

  • Memory

​

  • VLLMs

​

  • Loops

​

  • Human-in-the-loop

Ingredients

5 of 10

Agents

Graph State

Computational Graph

  • Task
  • Input
  • Output

Centralized register for messages

Used to manage communication among agents

Set agent context and memory

�

Set system workflow.

Can be static or dynamic

LangGraph

6 of 10

Start

Coder

executor

debugger

plotter

critic

human

End

7 of 10

https://github.com/franciscovillaescusa/AI-Science_Chicago_2026

https://tinyurl.com/5c2j7dv4

8 of 10

Discussion / Poll time

9 of 10

Discussion / Poll time

10 of 10

Discussion / Poll time