Bloomy bets an AI tutor with a hard 90% mastery gate can make tutoring affordable

What the business is

Bloomy is an AI-powered mastery-learning platform for K-12: it diagnoses each student's skill gaps, places them on a personalized learning path built on a knowledge graph of prerequisites, delivers standards-aligned lessons, and pairs practice with BloomyBot, a Socratic tutor restricted to the current lesson that scaffolds rather than gives answers. Each skill runs through three stages — Base Camp (taught lesson), Climb (guided practice) and Summit (independent 10-question assessment) — and students only advance at 90%+ mastery. Families subscribe (ELA $39/month or $279/year; Writing Studio $19/month or $139/year; Math launched July 31 at ELA pricing) and schools or microschools license per student.

Starting capital：YC Summer 2026 batch; backed by the founders of Clever per its YC page; no round size disclosed as of September 2026.

How it started

Alex Southmayd taught 7th-grade English and writing with Teach For America, where he struggled to differentiate for 30 students with 30 different needs; he then designed GMAT and GRE curriculum for Manhattan Prep, led Lyft's New England driver growth, earned a Stanford MBA, and ran AI transformation projects at McKinsey. When models got good enough in early 2026, he combined his learning-design, go-to-market and AI experience to build Bloomy, inspired by mastery-based schools like Alpha and by Benjamin Bloom's two-sigma finding that one-on-one tutoring massively outperforms classroom instruction.

What happened

Bloomy came out of stealth on 2026-07-20 with its Launch HN post and now operates across districts, charter schools, hybrid schools, microschools, homeschools and families. The product ships ELA, Writing and Math with a knowledge graph built in collaboration with Learning Commons / Chan Zuckerberg Initiative, multilingual BloomyBot support (Spanish, French and others) rolling out, and safety rails: conversation logging, no behavioral advertising, no model training on identifiable child data, and teacher or parent review of all tutoring activity.

How it ended up

Still early: revenue comes from family subscriptions and school licensing, and the company is testing whether AI tutoring beats static feedback in real classrooms — treating the 1.8x pilot result as an encouraging signal, not proof.

What can be applied

When a proven pedagogy is blocked by cost, attack the cost curve: keep the teaching loop rigorous and the mastery gate objective, and let AI replace expensive human attention, not learning design.