Build Your Own X: AI Edition

Project based learning has always been my gold standard for learning programming. It keeps things practical and grounded in building applications. Not just getting better at math and optimization such as with approaches like leetcode.

Examples of these repos include:

• practical-tutorials/project-based-learning
• karan/Projects
• Crafting Intepreters

Now in the age of AI, I think there are a lot of similar design patterns and applications that are constantly replicated. I’ll see the same playbook constantly.

1. Big lab releases a product
2. All the other labs race to copy it
3. Inevitably an open source alternative is released that outperforms the releases from big labs.
4. Repeat

So in that vein I think there are a set of different AI/agentic applications that we can build to learn the intuition of working with LLMs and common usecases.

Below I’ll dive into these different cases with links to examples of open implementations and tutorials.

---

Chatbots

• Chainlit
• open-webui

CLI Agents

• Amp — How to Build an Agent
• Announcing Toad - a universal UI for agentic coding in the terminal
• Yes, Claude Code can decompile itself. Here’s the source code.

Code Review Agents

• I built an AI code review agent in a few hours, here’s what I learned
• Kit Code Intelligence Toolkit

Deep Research

• Awesome Deep Research
• Open-source DeepResearch – Freeing our search agents
• ROMA: Recursive Open Meta-Agents

Co-pilots

• Co-pilot internals
• llm language server
• avante.nvim

Claude Artifacts or Agentic E2E Coding Agents

• Townie
• Open Lovable
• Freestyle