What you will build
By the end of this course, you will have a working personal AI operating system — a monorepo containing a public website, Python agents, MCP servers, and an observability stack.
This is not a theoretical course. Every lesson results in working code.
Who this is for
- Developers curious about AI agents who want to go beyond prompting
- Engineers who want to build their own AI-powered tooling
- Anyone who wants a structured, observable system for learning and building in public
You should be comfortable with Python and basic command-line usage. Some TypeScript helps for the web parts but is not required.
What you will learn
By the end of this course you will understand:
- How to structure a monorepo for a personal AI platform
- How to build Python agents with LangGraph
- How to use MCP (Model Context Protocol) to give agents controlled access to your tools
- How to set up Langfuse for observability and cost tracking
- How to make your system LLM-independent using LiteLLM
- How to publish your work as a live website
Course outline
Lesson 1: Architecture and mindset
Why a personal AI OS? The core loop. How to think about agents, tools, and content.
Lesson 2: Monorepo setup
Setting up pnpm workspaces, Turborepo, and a clean folder structure for the long term.
Lesson 3: The public website
Building the Next.js app with TypeScript, Tailwind, and MDX content. Getting live on Vercel.
Lesson 4: Content as files
Why file-based content is the right choice for agent-generated output. MDX frontmatter conventions. Reading and rendering content.
Lesson 5: Your first agent
Building a research agent with LangGraph. Tools, state, and the loop pattern. Connecting to a real source.
Lesson 6: MCP servers
Building a local MCP server to give your agent access to your content folder. Safety boundaries. Logging.
Lesson 7: LLM gateway and observability
Abstracting your LLM calls with LiteLLM. Setting up Langfuse. Understanding what your agents are actually doing.
Lesson 8: The full loop
Connecting everything: agent discovers → writes MDX → website publishes. Running it. Watching it in Langfuse.
Prerequisites
- Python 3.11+
- Node.js 20+
- Docker Desktop (for local Langfuse stack)
- A Claude, OpenAI, or Gemini API key (or Ollama running locally)
- Basic command line comfort
This course is in development. Sign up for updates by sending a note to hello@pranavsrivastava.com.