Learn2Earn AI Engineering Fellowship Python Project Series — Yekeen Maadan

A Python project series, one repo, built during the Learn2Earn AI Engineering Fellowship. Each project links to its README for the full writeup.

Across the first four projects the bugs weren't syntax. They came down to three questions: who owns the state, what type a value actually is, and when something runs. Going forward: type hints on function signatures, checked with pyright, starting in project 5.

• 01. Contact Book (CLI)

  Dictionaries, JSON persistence, separating validation from actions, ASSUMES / PROMISES / WILL NOT HANDLE

  • Assigned load_contacts instead of calling it, which caused a TypeError.
  • Used f.write() on a dict. It needed json.dump().
  • contact.json vs contacts.json: no error, just a second file and persistence that looked broken.

• 02. Study Planner Agent (AI agent (Groq))

  Calling an LLM API from a CLI, persisting chat history, reading an SDK instead of assuming it

  • Trailing commas turned strings into tuples, more than once.
  • Carried Gemini SDK assumptions into Groq and called methods on the wrong objects.
  • Missing commas between function parameters.

• 03. Invoice Generator (CLI + PDF (ReportLab))

  Classes that own their state, serializing objects to JSON, generating PDFs

  • Generated the PDF before the item loop finished, so it had no items.
  • Kept a separate items list instead of letting Invoice own its state.
  • Used range(num_of_items + 1) on a string from input(): wrong type and an extra iteration.

• 04. Sudoku (Desktop GUI (Flet))

  Recursive backtracking, puzzle uniqueness by counting solutions, redrawing the UI from state

  • Generator produced the same solved grid every game. Found in code review, not by testing.
  • Cell removal could loop forever when no more cells could be removed safely.
  • is_board_solved reported false conflicts by comparing each cell against itself.