IndustryK-12 Education / Family AI

All-Star Training Camp

A self-hostable AI family tutor that teaches 7 school subjects in a child’s own language.

In short

All-Star Training Camp is a self-hosted AI study coach for K-12 families. Parents hand-write a SOUL.md that defines the child's personality and interests, and the AI retells the same concept across seven school subjects — maths, English, Chinese and more — as a Minecraft, space or cartoon story, then checks real understanding with Socratic questions and a Feynman-style "teach it back". Its distinguishing trait is that it lives inside WeChat: photograph a handwritten problem and start, with no app to install and no new account.

What the agent takes off people's plates

Core capabilities

01

One persona per child

Parents hand-write a SOUL.md that defines the child; the AI retells the same concept as Minecraft / space / cartoon narratives.

02

Socratic + Feynman

Guided questions plus "teach it back" verification; error tracking generates daily micro-sprints.

03

Qwen vision model OCR

Photograph a handwritten problem → multimodal recognition → conversational re-phrasing.

04

WeChat native

No VPN, no app-switching — all interaction happens inside WeChat.

Side by side

An AI study coach vs general-purpose LLM chat vs a traditional question-bank app

DimensionTraditional question-bank appGeneral-purpose LLM chatAll-Star Training Camp
Does it know this particular childPushes the same grade-level content to every childEvery session starts from scratch; context must be re-explainedA parent-written SOUL.md persists; the same concept is retold in three narrative styles
Teaching methodDrill and read the explanation; whether it was understood stays unknownTends to hand over the answer, which invites copyingSocratic guiding questions plus a Feynman "teach it back" comprehension check
What happens to mistakesLogged in a mistake book, then it is on the parent to enforce reviewNo memory across sessions; yesterday's mistake is never revisitedError tracking generates a daily micro-sprint that comes back to it the next day
Entry point and data ownershipA separate app, with account and learning history held by the vendorA separate account, with conversation logs kept by the providerRuns inside WeChat, deployed per family — the learning data stays with that family

Subjects and capability matrix

Covered modules

Math

English

21,780 word corpus

Chinese

Biology

History

Civics

Geography

Verifiable facts

Numbers and sources

Seven subjects are covered: maths, English, Chinese, biology, history, civics and geography.

Self-reportedAll-Star Training Camp product configuration, verified 2026-082026-08-01

The English module ships with a 21,780-entry word corpus.

Self-reportedAll-Star Training Camp product configuration, verified 2026-082026-08-01

Any concept can be retold in three narrative styles — Minecraft, space or cartoon — with the parent-written SOUL.md deciding which one applies.

Self-reportedAll-Star Training Camp product configuration, verified 2026-082026-08-01

Handwritten problems are recognized by a Qwen-family vision-language model and re-framed as a conversation, so nothing has to be typed in.

Self-reportedAll-Star Training Camp product configuration, verified 2026-082026-08-01

Where it fits

When this approach does not apply

  • This is not an exam-score product. It trains understanding and articulation; the seven subjects are not aligned to any specific regional textbook edition or exam syllabus, so last-mile exam prep should still lean on school teaching and coursebooks.
  • Results depend heavily on the SOUL.md the parent writes. The more specific the persona, the closer the AI's narration lands; if a parent skips that investment, the experience degrades into ordinary AI chat.
  • Self-hosting means the family carries a server or home machine plus model-inference costs and routine maintenance. Families unwilling to take that on are better served by an off-the-shelf commercial product.
  • The AI can still be wrong. Definitions, formulas and exam-relevant conclusions need a parent or teacher to check, and a child should never treat the AI's answer as the only authority.
Written by: YGG Technology solutions teamPublished: 2026-05-08Last reviewed: 2026-08-01