Skip to content

Latest commit

 

History

History
174 lines (131 loc) · 3.2 KB

slides.md

File metadata and controls

174 lines (131 loc) · 3.2 KB
theme title transition
apple-basic
Learn Building AI Agents
fade

Learn LLM based Agents

By making things that have already been solved

By Brooks Patton

Code and slides: https://github.com/brooks-builds/meetup_talk_ai_todo

First a question

How many of you code?

  • doesn't matter what language
  • doesn't matter what framework

Objectives

  • To inspire you to create your own agents
    • without Langchain

AI (LLM) Agents

What do we mean by AI Agents anyways?

There isn't really a good definition. I've multiple descriptions of agents, from

  • thin wrapper around LLMS

to

  • app that has LLMs call tools

layout: image image: ./slide_images/demo_screenshot.png backgroundSize: contain

demo - A todo app

layout: image-right image: ./slide_images/high_level_overview.png backgroundSize: contain

How it works from a high level

Tech Stack

  • Ollama
  • Rust
  • Docker (Postgres Database)

The Key, tool calling

When using the API, we can tell the LLM that a function exists, what it takes in, and what it will return.

  • the LLM and the actual function are not intertwined, the LLM cannot interface with the code directly
  • The LLM can choose to call a function or not
  • Ollama based LLMs almost always call a function when given the chance to...even if it doesn't make sense

layout: two-cols

requesting a tool call from ollama
```json { "model": "qwen2:7b-instruct-fp16", "messages": [ { "role": "user", "content": "what is the weather like in Denver, Colorado?" } ], "stream": false, "tools": [ { "type": "function", "function": { "name": "get_weather", "description": "queries an api that will return the current temperature in Farenheight", "parameters": { "type": "object", "properties": { "location": { "type": "string", "description": "The location that you want to check the weather of. The format should be 'City, State'. for example 'Berkely, CA'" } } } } } ] } ```
::right::
response from ollama
{
	"model": "qwen2:7b-instruct-fp16",
	"created_at": "2024-12-04T22:33:07.605992Z",
	"message": {
		"role": "assistant",
		"content": "",
		"tool_calls": [
			{
				"function": {
					"name": "get_weather",
					"arguments": {
						"location": "Denver, CO"
					}
				}
			}
		]
	},
	"done_reason": "stop",
	"done": true,
	"total_duration": 2638946583,
	"load_duration": 563932916,
	"prompt_eval_count": 184,
	"prompt_eval_duration": 648000000,
	"eval_count": 27,
	"eval_duration": 1177000000
}

Challenge

Try creating your own agent without Langchain

  • start with using a tool like postman to learn tool calling
  • identify a simple problem that is already solved
  • create a very simple version of that solution but using LLMs as the core
  • don't worry about having the LLM do everything

Thanks for listening

Any Questions?

How to find me