Learning GenAI & Agentic AI

A hands-on learning repository for Cloud and DevOps engineers stepping into Generative AI and Agentic AI. Every concept is grounded in infrastructure analogies you already know.

---

Who Is This For?

You deploy containers. You write Terraform. You've been paged at 3am. Now AI is showing up everywhere in your stack — in CI/CD pipelines, incident response tooling, and the platforms your company is building.

This repo takes you from "I've used ChatGPT" to "I understand how this works and I can build with it."

Assumed background:

• Comfortable with Python (scripts, not data science)
• Hands-on experience with Kubernetes, Terraform, Docker, or similar cloud/DevOps tools
• Some AWS/GCP/Azure exposure helpful but not required

Not required:

• Math background
• Machine learning experience
• Data science background

---

Curriculum Map

#	Session	Key Tools	API Key?	Status
01	How LLMs Actually Work	tiktoken, sentence-transformers	No	✅ Available
02	Prompt Engineering	anthropic	Yes	✅ Available
03	Local LLMs with Ollama	ollama, requests	No (local)	✅ Available
04	Tool Calling (Function Calling)	openai, ollama	Depends on provider	✅ Available
05	MCP Servers with FastMCP	fastmcp, mcp	No	✅ Available
06	Memory Management in Agentic AI	Python standard library	No	✅ Available
07	Retrieval-Augmented Generation (RAG)	pypdf, tiktoken, sentence-transformers, chromadb, openai	Optional	✅ Available
08	Agent2Agent (A2A) Protocol	Python standard library	No	✅ Available
09	Fine-Tuning LLMs	Python standard library, openai optional	Optional	✅ Available
10	GenAI Evals and Red Teaming	Python standard library	No	✅ Available
11	LLMOps and Production GenAI	Python standard library	No	✅ Available
12	GenAI Security Frameworks	Python standard library	No	✅ Available
13	Modern Agent Runtime	Python standard library	No	✅ Available
14	Multimodal and Realtime AI	Python standard library	Optional	✅ Available
15	Open Model Production Serving	Python standard library	No	✅ Available
16	GenAI Observability	Python standard library	No	✅ Available
17	AI Governance, Risk, and Compliance	Python standard library	No	✅ Available

---

How Each Session Is Structured

sessions/XX_topic_name/
├── README.md          ← Start here: objectives, prereqs, estimated time
├── concepts/          ← The "why" and "how" — read before coding
│   ├── 01_topic.md
│   └── ...
├── labs/              ← Hands-on coding exercises
│   ├── lab01_name/
│   │   ├── lab.py     ← You write the code (has TODO markers)
│   │   └── solution.py ← Reference implementation (peek only when stuck)
│   └── ...
└── demos/             ← Ready-to-run scripts — observe and learn
    └── ...

How to use labs:

1. Open lab.py and read it top to bottom — understand the goal
2. Fill in the # TODO sections — use the concept docs if stuck
3. Run with python lab.py and verify your output matches the expected output in comments
4. If truly stuck after trying: check solution.py

---

Quick Start

# 1. Clone
git clone <repo-url>
cd LearningGenAI

# 2. Create a virtual environment
python -m venv .venv
source .venv/bin/activate        # Windows: .venv\Scripts\activate

# 3. Install dependencies
pip install -r requirements.txt

# 4. Verify your setup
python setup_check.py

# 5. Start Session 01
cd sessions/01_how_llms_work
cat README.md

Session 01 requires no API key. You learn tokenization and embeddings completely offline.

---

API Key Setup

For cloud-provider labs, copy and configure .env:

cp .env.example .env
# Open .env and fill only the keys you need:
# - ANTHROPIC_API_KEY (Session 02)
# - OPENAI_API_KEY / provider-compatible keys (Session 04)

Optional local path:

• Session 03 runs fully local with Ollama.
• Session 04 can also run locally through Ollama's OpenAI-compatible /v1 endpoint.
• Sessions 05, 06, and 08 do not require API keys for the included labs and demos.
• Session 07 core labs run without API keys; the optional generated-answer demo can use OPENAI_API_KEY.
• Session 09 core labs run without API keys; optional hosted fine-tuning examples use OPENAI_API_KEY.
• Sessions 10-17 core labs run without API keys. Optional extensions can connect to provider APIs, realtime APIs, OpenTelemetry collectors, or GPU model servers.

---

Repository Philosophy

• No notebooks. DevOps engineers live in terminals and editors. Every exercise is a plain .py script.
• Offline-first. Session 01 runs entirely without an internet connection after the initial pip install.
• DevOps analogies first. Every concept is introduced with an infrastructure analogy before the AI theory.
• Build things that matter. Labs produce code patterns (semantic search, context management, agent loops) you will actually use in production.

---

Contributing

Found a bug, a better analogy, or want to contribute a session? Open an issue or PR — contributions welcome.

---

License

MIT — use freely for learning and teaching.