Agentic AI Course : Week 1

8 weeks

https://edwarddonner.com/2024/11/13/llm-engineering-resources/

 6 week Agentic AI Course

https://github.com/ed-donner/agents

the file .env has all the keys

select kernel

https://github.com/ed-donner/agents/blob/main/1_foundations/1_lab1.ipynb

import os openai_api_key = os.getenv('OPENAI_API_KEY') from openai import OpenAI openai = OpenAI() messages = [{"role": "user", "content": "What is 2+2?"}] response = openai.chat.completions.create(     model="gpt-4.1-nano",     messages=messages     ) print(response.choices[0].message.content)

Day 2 : what are AI Agents?

5 different patterns for agentic systems

https://lavnish.blogspot.com/2025/05/essential-design-patterns-for-building.html

In General

Day 3

• comparing different models 
• https://www.vellum.ai/llm-leaderboard
• syntax to call diff LLMs openAI , anthropic, gemini, deepseek, groq
• https://github.com/ed-donner/agents/blob/main/1_foundations/2_lab2.ipynb 
• diff LLMs ans 1 ques, declare one winner

Day 4

Compare AI Agent

Resources vs Tools

• Resource 
• is a fancy way of saying you can improve an LLM with resources to improve its expertise
• the way you do that is grab relevant data and shove it into prompt
• there are techniques like RAG to get really smart at picking relevant content
• Tools
• power to do something at its own discretion
• Build a chatbot that acts like you
• https://github.com/ed-donner/agents/blob/main/1_foundations/4_lab4.ipynb 
• gradio is a framework to make UI
• has callback functions
• system prompt : is generally like "you are acting like a xxx"
• user prompt : is actual task
• You can host it like here as well https://huggingface.co/spaces/ed-donner/Career_Conversation 
• Using gemini to evealuate GPT4 response : Multi LLM pipeline
• use another LLM to evaluate an answer
• Pydantic is a Python library for data validation, data parsing, and serialization that uses Python's type hints to define data schemas, making code more robust, readable, and easier to debug.
  

  
  
• check line 11 https://github.com/ed-donner/agents/blob/main/1_foundations/3_lab3.ipynb 
1. We built a function line 27 that uses "gpt4" to ans questions
2. then line 29 to evaluate answer using gemini , using evaluation object
3. final piece of puzzle is use a rerun function 
• this ( chat function , line 35 ) is an example of simple workflow without any framework
• 
  

Day 5

• globals() : https://www.geeksforgeeks.org/python/python-globals-function/
• https://github.com/ed-donner/agents/blob/main/1_foundations/4_lab4.ipynb
• main thing is function handle_tool_calls()
• you can re write app like this, and you can deploy using hugging face spaces.
• command to deploy is "gradio deploy"
• 
  

Misc

https://highlight.hohli.com/?language=python