Hey dev community! 👋 **
I'm Zkzk from Cairo, and I've been deep in agentic AI lately. After struggling to find a clear, hands-on path from "hello LLM" to "production-ready autonomous agent," I built my own tutorial repo: **Agentic-AI-Tutorial.
It's a 5-chapter progression (4 complete, Chapter 5 launching soon) with executed Jupyter notebooks, local Ollama support, OpenAI/Gemini options, and a real-world capstone: a Personal Finance Tracker Agent that tracks expenses, analyzes patterns with RAG, and gives smart advice via FastAPI + ChromaDB.
If you're in Cairo or anywhere grinding on AI agents in 2026, this is for you. Let's break it down!
Why Agentic AI Matters Right Now
LLMs are powerful, but they're stateless chatbots without memory or tools.
Agentic AI changes that: agents reason, use tools, remember context, self-refine, and even collaborate.
My tutorial shows you how — practically, reproducibly, and locally-first (privacy + zero cost with Ollama).
Chapter Breakdown: The Journey
Chapter 1: LLM Foundations
Call models from Ollama (local Gemma/Qwen), OpenAI (gpt-4o-mini), Gemini.
Learn streaming, system prompts, token counting.
Notebook highlights:
- Simple chat
- Emoji-styled responses
- Multi-provider switching
Great for beginners testing local vs. cloud.
Chapter 2: LangChain Chains & Tools
Build interactive workflows with LCEL.
Add memory (RunnableWithMessageHistory), bind tools, create routers/parallel chains.
Demo: Stateful chat that remembers your name + routes queries dynamically.
Chapter 3: Memory + RAG Pipelines
Persistent memory types + local embeddings (sentence-transformers).
RAG setup with ChromaDB prep (full vector store coming in Ch5).
Focus: Cost/privacy with everything local.
Chapter 4: LangGraph Orchestration (My Favorite!)
Move from chains to graphs.
Build ReAct agents, multi-agent collab, self-refinement loops.
Notebook magic:
- Graph visualization with Graphviz (nodes/edges appear live!)
- State updates in real-time
- Tool calls (e.g., math/search)
- Conditional routing + human-in-the-loop
This chapter feels "alive" — agents actually think!
Chapter 5: Real-World Capstone – Personal Finance Tracker Agent
Tying it all together:
- FastAPI backend for API endpoints (/track_expense, /query_budget)
- ChromaDB for embedding/storing transactions + finance tips (local RAG)
- LangChain chains for parsing/categorizing/calculating
- LangGraph for orchestration: Parse → Retrieve → Analyze → Advise (ReAct + refinement)
- Stateful memory per user/session
Example flow:
User: "Groceries at Carrefour, 500 EGP"
Agent: Queries patterns → "Your food spending up 20% this month – try these local deals!"
Deploy locally with uvicorn, ready for extensions (Streamlit UI, EGP exchange rates).
Tech Stack (All Reproducible)
- Python 3.10+
- LangChain, LangGraph
- Ollama (local), OpenAI, Google Gemini
- ChromaDB + sentence-transformers
- FastAPI + uvicorn
- .env for keys, venv for isolation
No paid walls — run everything locally!
Quick Start
- Clone:
git clone https://github.com/zkzkGamal/Agentic-AI-Tutorial -
cdinto folder →python -m venv venv→ activate pip install -r requirements.txt- Set up .env (API keys if using cloud)
- Run notebooks in order (Chapter1 → Chapter5)
What's Next?
Chapter 5 notebook + deployment guide dropping soon.
Planning extensions: Streamlit frontend, more tools (e.g., currency API), multi-user persistence.
Call to Action
If this resonates:
- ⭐ Star the repo – it really helps!
- 🍴 Fork & build your own agent (maybe a Cairo event planner next?)
- Comment below: What real-world agent would YOU build with these tools?
Repo: https://github.com/zkzkGamal/Agentic-AI-Tutorial
Let's build agentic AI together
Top comments (0)