Dare2Solve Studio: fresh math puzzles a tutor can trust, written by a model on their own laptop

Описание и применение

Dare2Solve Studio генерирует новые математические задачи для учителей, проверяя ответы с помощью кода и повторного решения, хранит их в дублирующейся базе и предоставляет готовые рабочие листы.

Задача

Tutors in India need fresh math puzzles weekly but existing sources are either paid, repetitive, or unreliable because chatbot answers may be wrong.

Результат

In a test run across eight topics, all eight puzzles passed checks and five of thirteen attempts were discarded, demonstrating effective validation. The system runs locally on a laptop with 6 GB GPU, keeping costs near zero and preserving privacy.

Рабочий процесс

• 1. Tutor selects topic, audience, and difficulty.
• 2. Model (Gemma 4 E4B via Ollama) generates puzzle JSON with problem, answer, worked solution, and a Python solve() function.
• 3. Code check runs solve() in a sandbox to confirm answer matches output.
• 4. Blind re-solve has model re-solve the problem text independently to catch inconsistencies.
• 5. If all checks pass, puzzle is stored in Tiger Cloud bank; otherwise it is discarded.
• 6. Tutor retrieves puzzles from bank, prints worksheet PDF with answer key or creates a square card for WhatsApp/Instagram.

Доказательства

Dare2Solve Studio writes fresh math puzzles and refuses to hand over any puzzle whose answer it couldn't confirm.

Источник

The model writes five things for each puzzle: the problem, its working notes, a Python solve() function that computes the answer independently (by brute force where possible), a worked solution, and the answer.

Источник

I ran the pipeline on eight topics, one puzzle each, up to three attempts per puzzle, with Gemma 4 E4B: 8 of 8 topics produced a puzzle that passed every check, and 5 of 13 attempts were discarded on the way (about 30 minutes on the laptop).

Источник

Verified puzzles go into a Tiger Cloud hypertable with a 384-dimensional embedding of the problem text (bge-small-en-v1.5 via fastembed, also local) and an HNSW index.

Источник