Hint Tutor gives offline Python bug hints without revealing the fix

Hint Tutor is an open-source tool for introductory Python courses that returns hints pointing to a student's bug without supplying the corrected code. It runs fully offline on a 3-billion-parameter model through Ollama, and unit tests decide first whether the code is wrong, so the model is only called to explain failing tests. A three-stage pipeline localizes the faulty line, drafts a short hint ending in a question, and checks every draft against the reference solution to block code or revealing tokens, falling back to a generic template after three failed attempts. On a 50-case benchmark, the system identified the exact faulty line 60% of the time and within one line 74% of the time, though it matched bug categories only 42% of the time. The author notes that testing was largely synthetic, that the guard does not catch English paraphrases of the fix, and that the tool only handles single-file scripts under 150 lines without input.