convez.ai
MCP Server
bipinhcs11

Agentic-AI-Learning-Roadmap

bipinhcs11/Agentic-AI-Learning-Roadmap

Ver no GitHub
InicialPluginScore · de 0 a 100

Local-first hands-on roadmap for learning Agentic AI, RAG, MCP, multi-agent systems, and production AI apps.

Instalação

$npx convez-ai bipinhcs11/Agentic-AI-Learning-Roadmap

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
2
Forks
0
Acompanhando
0
Saúde
57%
Último commit
há 1 mês
Versão
—
Issues abertas
0
Contribuidores
1

README

A tradução para o português deste README ainda está na fila.

Agentic AI Learning Roadmap 🤖

Local-first hands-on roadmap for learning Agentic AI, RAG, multi-agent systems, MCP, and production AI apps with Ollama, LangGraph, CrewAI, FastAPI, and AWS deployment patterns.

9 phases | 48+ projects | suggested 48+ week pace | 100% local-first

Origin: This roadmap started with a viral LinkedIn post — "Build these 10 RAG projects if you want to be taken seriously as an AI engineer." Those 10 projects became Phase 2, and the journey grew from there into a full 9-phase path: foundations → RAG → agents → frameworks → multi-agent → production → advanced patterns → shipping → dynamic MCP enterprise systems.

This repo is built for two audiences:

  • Self-paced builders who want to learn by running real projects locally.
  • Community learners who want a practical map from first local model to production-style agentic systems.

The week numbers are suggested pacing, not a rigid calendar. Move faster, slow down, or jump to the phase that matches what you are trying to build.


Start Here

If you are...Start withWhy
New to local AIPhase1_Foundations/Get Ollama, Python, and the local OpenAI-compatible client working.
Learning RAGPhase2_RAG_Systems/project_01_first_rag/Build the first retrieval pipeline from scratch before frameworks.
Learning agentsPhase3_Agentic_Stack/Move from tool calling to memory, scraping, and API agents.
Learning production AIPhase6_Production_Enterprise/Docker, auth, observability, deployment, and capstone SaaS patterns.
Learning MCPPhase9_Dynamic_Agentic_RAG_MCP/MCP tools, MCP + RAG routing, and the enterprise assistant hub capstone.

Learning Paths

You do not have to complete everything in order.

  • Beginner path: Phase 1 → Phase 2 Project 01 → Phase 3 Project 01
  • RAG engineer path: Phase 2 → Phase 3 Project 05 → Phase 7 Project 01
  • Agent builder path: Phase 3 → Phase 5 → Phase 7
  • Production path: Phase 6 → Phase 8 → Phase 9 capstone
  • MCP path: Phase 9 Module 01 → Module 02 → Enterprise Assistant Hub

The Stack

LayerTechnology
Local AIOllama + Gemma3:4b / 27b
OrchestrationLangGraph, CrewAI, LangChain
Vector DBChromaDB → Qdrant → Pinecone
APIsFastAPI + Uvicorn
UIStreamlit
Multi-AgentLangGraph StateGraph, CrewAI Crews
ContainersDocker + Docker Compose
CloudAWS ECS Fargate + ECR + ALB
IaCTerraform
MonitoringPrometheus + Grafana
Fine-TuningLoRA + Unsloth + GGUF
IntegrationsSlack, GitHub, Stripe, Email
MCPModel Context Protocol tools, resources, prompts

Phases

Phase 1 — Foundation (Weeks 1-2) ✅

Goal: Get local AI running on Mac Mini M4

  • Installed Ollama, Python 3.11, VS Code
  • OpenAI-compatible local API client
  • Verified full stack with gemma3:4b

📁 Phase1_Foundations/


Phase 2 — RAG Projects (Weeks 3-6) ✅

Goal: Build the 10 hands-on Retrieval-Augmented Generation projects from the original LinkedIn post

#Project
01First RAG pipeline (build from scratch)
02IBM-style RAG (production patterns)
03GraphRAG (knowledge graph)
04Multi-document RAG (vector database)
05Agentic RAG (autonomous agents)
06LangChain RAG (production ready)
07Document analysis (LLM + PDF)
08Multimodal RAG (text + images)
09AI research agent (automated analysis)
10Real-time assistant (live RAG pipeline)

Stack: nomic-embed-text, ChromaDB, numpy cosine similarity

📁 Phase2_RAG_Systems/


Phase 3 — Full Agentic Stack (Weeks 7-12) ✅

Goal: Master the complete agentic AI tech stack

#Project
01Tool-calling agent (ReAct pattern)
02Memory agent (short + long term)
03Web scraping agent
04Multi-tool agent
05RAG evaluation (LLM-as-judge)
06Agent API server

Stack: FastAPI, mem0, BeautifulSoup, Ragas

📁 Phase3_Agentic_Stack/


Phase 4 — Build Your Own Agent Framework (Weeks 13-16) ✅

Goal: Build something similar to LangChain from scratch

#Project
01Model manager
02Inference server (streaming + logging)
03OpenAI-compatible API
04Streamlit web UI
05Custom agent framework (mini LangChain)
06Full platform capstone

Stack: FastAPI, Streamlit, SQLite, Typer

📁 Phase4_Agent_Framework/


Phase 5 — Multi-Agent Systems (Weeks 17-22) ✅

Goal: Multiple specialized agents coordinating to solve complex tasks

#Project
01Supervisor-Worker pattern (LangGraph)
02CrewAI Research Crew (Researcher → Analyst → Writer)
03Agent Communication Bus (asyncio pub/sub)
04Code Generation Pipeline (review-revise loop)
05Multi-Agent RAG with domain routing
06Autonomous Research Pipeline with human-in-the-loop

Stack: LangGraph, CrewAI, asyncio, Redis (optional)

📁 Phase5_Multi_Agent_Systems/


Phase 6 — Production & Enterprise (Weeks 23-30) ✅

Goal: Take everything and make it production-ready, observable, secure, and deployable

#Project
01Dockerize Everything (API + UI + Nginx)
02Auth & RBAC (JWT + roles + API keys)
03AWS Deployment (ECS Fargate + Terraform)
04Observability (Prometheus + Grafana dashboards)
05Fine-Tuning Gemma3:4b on Apple Silicon M4
06DocuMind — AI document intelligence SaaS (capstone)

Stack: Docker, Terraform, AWS ECS, Prometheus, Grafana, LoRA, Unsloth

📁 Phase6_Production_Enterprise/


Phase 7 — Advanced AI Patterns (Weeks 31-36) ✅

Goal: Go beyond basic agents into cutting-edge production patterns

#Project
01GraphRAG (knowledge graph + relationship traversal)
02Real-time streaming (WebSocket token streaming)
03Long-term memory (persistent vector memory)
04Mixture of Agents (query routing to specialists)
05Self-improving agent (Reflexion loop)
06AI safety & red-teaming (guardrails + adversarial tests)

Stack: networkx, FastAPI WebSockets, SQLite, LangGraph

📁 Phase7_Advanced_AI_Patterns/


Phase 8 — Integrations & Shipping (Weeks 37-44) ✅

Goal: Take local-AI into real-world integrations and ship a SaaS

#Project
01Slack bot (Socket Mode)
02GitHub review bot (PR webhook → AI review)
03Email agent (classify, prioritize, draft)
04Multi-tenant SaaS (FastAPI + JWT + quotas)
05Billing & metering (token usage + invoices + Stripe)
06Capstone launch (dockerized RAG SaaS)

Stack: slack-bolt, FastAPI, SQLAlchemy, Stripe, Docker

📁 Phase8_Integrations_Shipping/


Phase 9 — Dynamic Agentic RAG + MCP (Weeks 45+) 🔄

Goal: Learn MCP as its own enterprise integration pattern, then combine it with RAG and dynamic LLM provider routing

#Module
01MCP Benefits Assistant — mock 401(k) + HSA tools/resources
02MCP + RAG Enterprise Integration — structured tools + document retrieval
03Enterprise Assistant Hub capstone — MCP gateway + RAG + AWS provider abstraction
04Java MCP Benefits Assistant — plain Java companion to Module 01
05Spring Boot MCP + RAG Benefits Microservice — Streamable HTTP MCP + REST

Stack: MCP Python SDK, FastMCP, Spring AI MCP, Java, Spring Boot, local mock data, Ollama, NumPy, Bedrock/SageMaker adapters

📁 Phase9_Dynamic_Agentic_RAG_MCP/


Quick Start

# Clone
git clone https://github.com/bipinhcs11/Agentic-AI-Learning-Roadmap.git
cd Agentic-AI-Learning-Roadmap

# Setup Python environment
python3 -m venv ai-env
source ai-env/bin/activate
pip install -r requirements.txt

# Start Ollama
ollama pull gemma3:4b
ollama serve

# Run any project
python Phase2_RAG_Systems/project_01_first_rag/rag_from_scratch.py

Community

This started as a personal learning roadmap, but it is open for people who want to follow along, reuse examples, or suggest improvements.

  • Star the repo if you are following the roadmap.
  • Open an issue if a setup step is unclear or a project breaks.
  • Use GitHub Discussions for roadmap questions, project ideas, and learning notes.
  • See CONTRIBUTING.md for the contribution style.

Run the Capstone Product (DocuMind)

cd Phase6_Production_Enterprise/project_06_capstone_product
docker compose up --build
python demo/seed_data.py
# Open http://localhost  →  login: admin / admin123

Hardware

All projects run locally on:

  • Mac Mini M4 (16GB unified memory)
  • No cloud GPU required for any phase including fine-tuning
  • Ollama handles model serving natively on Apple Silicon

Repository Structure

├── Phase1_Foundations/             # setup + first model (docs/, test_gemma3.py)
├── Phase2_RAG_Systems/           # 10 RAG projects (+ guide)
├── Phase3_Agentic_Stack/          # 6 agent projects (+ guide)
├── Phase4_Agent_Framework/        # 6 framework projects (+ guide)
├── Phase5_Multi_Agent_Systems/    # 6 multi-agent projects
├── Phase6_Production_Enterprise/  # 6 production projects
├── Phase7_Advanced_AI_Patterns/   # 6 advanced-pattern projects
├── Phase8_Integrations_Shipping/  # 6 integration / shipping projects
├── Phase9_Dynamic_Agentic_RAG_MCP/ # MCP benefits, MCP+RAG, enterprise capstone
├── scripts/                       # setup & install helper scripts
├── requirements.txt               # shared dependencies
├── CLAUDE.md                      # project context
└── README.md
Ver o original no GitHub

Bom para

Relacionados