Agentic AI Roadmap 🤖

A 12-week, self-directed curriculum for learning to build production-grade AI agents from first principles — no bootcamp, no course, just a structured sequence of daily builds using a fully local stack.

This repo is both my learning log and a template. If you want to run the exact same roadmap yourself, everything you need to replicate it is below.

Why this exists

Most "learn AI agents" content is either a shallow LangChain tutorial or a paid course that abstracts away the parts worth struggling with. This roadmap is built around daily hands-on files — each day produces working code, not notes. The goal is a portfolio of small, real artifacts (not one polished capstone) plus the muscle memory to build agentic systems from scratch, entirely on local/open-source tooling.

Stack

• Language: Python 3.11.9 (via pyenv)
• LLM runtime: Ollama — fully local, no API costs
  • qwen2.5:7b — primary chat/reasoning model
  • nomic-embed-text — embeddings for RAG
  • qwen2.5:14b — optional larger backup model
• Agent frameworks: LangGraph, CrewAI
• Vector DB: ChromaDB
• API layer: FastAPI
• Validation: Pydantic
• Protocol: MCP (Model Context Protocol)
• Deployment: Docker
• Client: OpenAI SDK (Ollama exposes an OpenAI-compatible endpoint — no vendor API key needed)

No paid API keys are required to complete this roadmap. Everything runs on your own machine.

Curriculum

| Weeks | Focus | |---|---| | 1–2 | Python fundamentals for agent codebases + Git workflow | | 3 | LLM APIs, prompt engineering, structured outputs, streaming | | 4 | Function calling, ReAct loops, task decomposition, embeddings, intro RAG + ChromaDB | | 5 | LangGraph fundamentals — nodes, edges, state | | 6 | MCP protocol + FastAPI | | 7 | CrewAI | | 8 | Multi-agent systems + Composio | | 9 | Memory systems + advanced/production RAG | | 10 | Vector databases at production scale | | 11 | Deployment + Docker | | 12 | Observability, testing, portfolio polish |

Each week produces its own set of files (e.g. week3_project.py), plus a WEEKX_NOTES.md and WEEKX_SUMMARY.md documenting what was learned and what was hard. This is the part worth not skipping — the notes are what make revision possible later.

Repo structure

agent_foundations/
├── .python-version          # pyenv local version (3.11.9)
├── requirements.txt          # all Python dependencies
├── README.md
│
├── week1_*.py ... week12_*.py    # daily/weekly build files
├── WEEKX_NOTES.md                # per-week concept notes
├── WEEKX_SUMMARY.md              # per-week retrospective
│
├── src/                      # project-structure practice
├── outputs/                  # file outputs from exercises
├── research_results/         # research/RAG outputs
├── sample_texts/             # practice input files
└── chroma_db/, week4_agent_db/   # ChromaDB storage (gitignored if large)

Replicate this roadmap

1. System prerequisites

sudo apt update
sudo apt install -y build-essential libssl-dev zlib1g-dev \
libbz2-dev libreadline-dev libsqlite3-dev curl git \
libncursesw5-dev xz-utils tk-dev libxml2-dev libxmlsec1-dev \
libffi-dev liblzma-dev

2. Install pyenv + Python 3.11.9

curl https://pyenv.run | bash

echo 'export PYENV_ROOT="$HOME/.pyenv"' >> ~/.bashrc
echo 'export PATH="$PYENV_ROOT/bin:$PATH"' >> ~/.bashrc
echo 'eval "$(pyenv init -)"' >> ~/.bashrc
source ~/.bashrc

pyenv install 3.11.9

3. Clone and set up the project

cd ~
git clone https://github.com/Aniruddhahrid/agentic-ai-roadmap.git agent_foundations
cd agent_foundations

pyenv local 3.11.9
python -m venv .venv
source .venv/bin/activate

pip install -r requirements.txt

4. Install Ollama and pull models

curl -fsSL https://ollama.com/install.sh | sh

ollama pull qwen2.5:7b
ollama pull nomic-embed-text

ollama list   # confirm both models are present

5. Verify everything works

python -c "import openai, pydantic, chromadb, langgraph; print('✓ All imports successful')"

python -c "
from openai import OpenAI
c = OpenAI(base_url='http://localhost:11434/v1', api_key='ollama')
r = c.chat.completions.create(model='qwen2.5:7b', messages=[{'role':'user','content':'hi'}])
print('✓ Ollama working:', r.choices[0].message.content[:50])
"

If both checks pass, you're ready to start Day 1.

How to actually run the roadmap

• One file per session. Each day/week's file is meant to be worked through, not just read — type it out, break it, fix it.
• Write the notes file before moving on. WEEKX_NOTES.md — what clicked, what was hard, in your own words. This is the single highest-leverage habit in the whole roadmap; skipping it is why revision later is painful.
• Commit daily. Small, frequent commits create a record you can actually revisit:

  git add .
  git commit -m "day X: description"
  git push
  

• Don't skip ahead. Weeks build on each other — Week 5 (LangGraph) assumes Week 4's function-calling and ReAct concepts are solid, Week 9 (memory/advanced RAG) assumes Week 4's intro RAG is second nature.

A note on local-first

Every model call in this roadmap runs through Ollama on localhost:11434, using the OpenAI SDK's compatibility layer:

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:11434/v1",
    api_key="ollama"  # dummy key, Ollama doesn't check it
)

This means the whole roadmap works offline, with zero API spend, on a machine with ~16GB+ RAM (32GB comfortable for qwen2.5:14b if you want the larger backup model). If you'd rather use a hosted API (OpenAI, Gemini, Groq, etc.), the code should port over with minimal changes — just swap base_url/api_key — but the roadmap as designed doesn't require it.

Status

Progress tracker (update as you go):

• [x] Weeks 1–2 — Python fundamentals + Git
• [x] Week 3 — LLM APIs, prompt engineering, structured outputs
• [x] Week 4 — Function calling, ReAct, RAG basics, ChromaDB
• [ ] Week 5 — LangGraph
• [ ] Week 6 — MCP + FastAPI
• [ ] Week 7 — CrewAI
• [ ] Week 8 — Multi-agent systems + Composio
• [ ] Week 9 — Memory + advanced RAG
• [ ] Week 10 — Vector databases
• [ ] Week 11 — Deployment + Docker
• [ ] Week 12 — Observability + portfolio

Questions / replicating this yourself

Open an issue or fork the repo. If you're doing this alongside me, I'd genuinely like to compare notes — ping me.