How techloset/agentic-ai works
Project Type and Purpose
• This repository is a monorepo for a “Cloud-Native Applied AI Agentic Developer” training course.
• It is not a single product but a collection of progressively organized modules and example code covering:
– Python foundations
– AI theory
– FastAPI back-ends
– Database integrations
– Data preprocessing
– Model development
– LLM and prompt engineering
– Agent frameworks (OpenAI Agent SDK, LangChain, LangGraph, CrewAI)
– Deployment (Docker, devcontainer)
– A Next.js front-end scaffold
File Structure
• Top-level directories represent logical layers or modules in the course pipeline:
– 01_python: basic Python scripts & readme
– 02_ai_theory: theory notes (Markdown)
– 03_fastapi: FastAPI demos (hello, routing, middleware, file handling, exceptions) and a “folder_standard” scaffold with routes/controllers/utils
– 04_databases: examples with MongoDB and Postgres (SQLAlchemy/Alembic migrations, controllers, models, routes, utils, validations)
– 05_data_preprocessing: NumPy, Pandas, OpenCV, NLTK scripts and sample CSVs
– 06_model_development: model training (MNIST, cat/dog), prediction scripts, FastAPI prediction endpoint
– 07_llm_and_prompt_engineering: basic LLM calls, settings, prompting techniques
– 08_openai-agent-sdk: an SDK plus example agents, tools, streaming, contexts, guardrails, orchestration
– 09_langchain: LangChain chains, memory, tools, RAG examples
– 09a_langgraph: LangGraph agent examples, FastAPI integration, memory backends, WebSocket agents
– 10_crewai: CrewAI flow examples, knowledge agents
– 11_deployment: Dockerfile
– 12_frontend: Next.js boilerplate
– assignments, data, books, main.py, pyproject.toml
Readme Insights
• Explains using Poetry for dependency management and virtual environments
• Shows how to run Python scripts and serve FastAPI apps with Uvicorn
• Contains a “Course Outline” flowchart:
Python → AI Theory → FastAPI → Databases → Third-party libraries → Model Dev → LLMs → LangChain → LangGraph → CrewAI → Cloud/DevOps → Frontend
a. Developer Environment
• Poetry / pyproject.toml
• .devcontainer (VS Code Development Container)
b. Core Language Layer
• 01_python
• 02_ai_theory (documentation)
c. API Layer (Backend Services)
• 03_fastapi (folder_standard app)
• 06_model_development/fastapi_prediction_model.py
d. Data Layer
• 04_databases/Postgres (SQLAlchemy + Alembic)
• 04_databases/MongoDB
e. Data Pipeline
• 05_data_preprocessing (NumPy, Pandas, OpenCV, NLTK)
f. ML Model Layer
• 06_model_development (training & prediction artifacts)
g. LLM & Prompt Layer
• 07_llm_and_prompt_engineering
• External LLM APIs (OpenAI, Gemini)
h. Agent Frameworks (Orchestrator Layer)
• 08_openai-agent-sdk
• 09_langchain
• 09a_langgraph
• 10_crewai
i. Front-End Layer
• 12_frontend (Next.js scaffolding)
j. Deployment & DevOps
• 11_deployment/Dockerfile
• uv.lock, .env.example, Docker, Kubernetes (implied)
Relationships and Interactions
• Frontend ↔ API Layer (HTTP/REST via Uvicorn)
• API Layer ↔ Data Layer (ORM calls to Postgres, drivers to MongoDB)
• API Layer ↔ Data Pipeline (invokes preprocessing scripts)
• API Layer ↔ ML Model Layer (loads model artifacts for inference)
• API Layer ↔ LLM & Prompt Layer (calls external OpenAI/Gemini endpoints)
• Agent Frameworks orchestrate calls across API, Data, Models, LLMs
• Deployment containerizes API, Agent services, and Frontend
Architectural Patterns & Principles
• Modular monorepo with clearly separated concerns
• Layered architecture (presentation, API, data, model, orchestration)
• Use of standardized FastAPI code scaffolding (routes/controllers/utils)
• Database migrations via Alembic (schema versioning)
• Dependency management via Poetry
• DevContainer for consistent developer environment
• Micro-examples that can be combined into composite microservices