Alex Southmayd, a former 7th-grade English and writing teacher with Teach For America, has launched Bloomy, an AI-powered learning platform for K-12 students at bloomylearning.com. The platform covers Math, English Language Arts, and Writing, and includes an AI tutor called BloomyBot, built on a mix of Anthropic and OpenAI models. The goal is direct: solve a famous education challenge known as the Bloom 2-sigma problem using AI (Hacker News).

In 1984, educational psychologist Benjamin Bloom found that students who received one-on-one tutoring with mastery-learning methods performed two standard deviations better than students in traditional classrooms. To put that in perspective, the average tutored student outperformed roughly 98% of students in a regular class. The education technology world has been trying to close that gap at scale for decades. Bloomy's approach is to use AI not to give answers but to act as a Socratic tutor — asking questions that guide students toward finding the answers themselves.

Each skill on Bloomy follows three stages. Base Camp is where students study worked examples. Climb is a practice phase where BloomyBot offers guidance by asking questions. Summit is a test with no hints and no AI help. Students must score at least 90% on the Summit to move on to the next skill. Keeping the tutor separate from the test is a deliberate choice: it stops the AI from making students look more capable than they actually are by helping during the assessment.

BloomyBot adapts its help based on how the student is doing. Rather than jumping in with a heavy hint right away, it first asks what the student tried, then increases support if needed. This matches how good human tutors work — giving carefully timed hints rather than just telling the student the answer.

The curriculum was built with help from Learning Commons, a partner of the Chan Zuckerberg Initiative. Bloomy works with outside assessments and also gives its own diagnostic test to create a personalized learning path for each student. Support for Spanish, French, and other languages has started rolling out.

Southmayd has pointed to Alpha School as an inspiration. Alpha School advances students when they show they have mastered a skill, rather than when they have spent a certain number of hours in a seat. Bloomy applies the same idea through its three-stage system and the 90% Summit requirement.

Families can sign up at bloomylearning.com/families, and a product demo is available at https://youtu.be/XHvoKt6qMeo.

The broader context is that since powerful AI language models became widely available, many edtech products have focused on generating content or grading assignments automatically. Bloomy's focus on question-based tutoring within a mastery-learning framework is a different approach. Whether an AI tutor can provide the kind of patient, adaptive guidance that human tutors give — and that produces those dramatic learning gains — is still an unanswered question. The design, though, lines up with what education research says works: separating teaching from testing, requiring mastery before moving on, and adjusting help based on how the student struggles.

The decision to use models from both Anthropic and OpenAI suggests the team picks the best model for each task rather than relying on a single provider.

The platform is still early in its rollout. Whether Bloomy delivers on its ambitious goal will depend on things no software design can promise: how good the curriculum map is, how well the tutoring adjusts to different types of students, and whether the quality holds up as it expands across subjects and languages. But the choices built into the product show real engagement with education research, not just a chatbot bolted onto a website. That alone sets it apart from much of what has entered the K-12 AI tutoring space since 2023.