The Center for Data Innovation recently spoke with Alex Southmayd, founder of Bloomy, a New York-based company developing an AI-powered learning platform for K–12 education. Southmayd explained how Bloomy uses an AI system to deliver personalized instruction, guide students through one-on-one tutoring, and provide teachers with insights into students’ learning.
David Kertai: What problem is Bloomy solving?
Alex Southmayd: Student achievement has been declining for more than a decade, even as schools invest heavily in educational technology. At the same time, decades of research have shown that individualized instruction is one of the most effective ways to improve learning. As a former seventh-grade English Language Arts teacher, I saw that challenge firsthand. Thirty students moved through the same lesson at the same pace, even though some lacked foundational skills while others were ready to move ahead. Later, as a tutor, I saw the opposite. When you identify exactly what a student does not understand and help them master it, progress can be remarkable. The challenge has always been scale. Schools cannot realistically provide every student with a dedicated tutor.
Bloomy addresses this challenge by using large language models to provide the benefits of one-on-one instruction at scale. The platform delivers personalized, mastery-based learning in English Language Arts, math, and writing. Within Bloomy, BloomyBot serves as an AI-powered tutor that guides students through individualized learning paths by asking questions, providing feedback, and helping students work through challenges rather than simply giving answers. Together, Bloomy’s assessments, instruction, and teacher insights create a classroom-ready platform that makes personalized learning accessible to every student.
Kertai: How does Bloomy create an individualized learning experience for each student?
Southmayd: Every learning path begins by understanding where a student actually is rather than simply what grade they are in. Bloomy runs its own diagnostic assessment, and teachers can also upload results from benchmark assessments, such as NWEA MAP and i-Ready, allowing schools to build on information they already have.
Students then work through roughly 30-minute learning modules and advance only after demonstrating mastery by scoring at least 90 percent on an assessment. Throughout each lesson, BloomyBot uses a Socratic approach, asking guiding questions and providing scaffolding instead of giving away answers. This keeps students engaged in productive struggle while allowing instruction to adapt to their individual needs.
Kertai: How does your AI system determine what each student should learn next?
Southmayd: Bloomy uses a knowledge graph built on established educational standards from Learning Commons and Illustrative Mathematics to map how skills connect and which concepts students need to master first. Rather than simply recognizing an incorrect answer, the platform identifies the underlying cause of a student’s difficulty.
For example, if a student struggles to identify an author’s purpose, the underlying issue may be difficulty understanding complex sentence structure. Instead of repeatedly practicing the same skill, Bloomy directs the student to strengthen that prerequisite first. BloomyBot also remembers previous interactions, allowing it to recognize recurring misconceptions and adapt its guidance based on how each student learns over time.
We take a deliberate approach to personalization. Rather than generating instructional content on demand, Bloomy uses preauthored lessons, reducing hallucination risks and giving educators control over instructional quality. The platform personalizes instruction by adjusting difficulty and reading complexity while recognizing when challenges in one subject, such as literacy, may affect performance in another, such as mathematics.
Kertai: How do you ensure your AI system provides accurate feedback?
Southmayd: Accuracy begins with the platform’s architecture. BloomyBot is grounded in preauthored lesson content, stays focused on the assigned material, and operates separately from student assessments. Bloomy measures mastery through a ten-question assessment completed without hints or BloomyBot assistance. Students must score at least 90 percent before advancing.
Bloomy also evaluates every BloomyBot response through an automated safety system before it reaches the student. A second independent review analyzes stored conversations using rule-based checks and a separate AI model to identify crisis situations, distress, or inappropriate content across multiple languages. When necessary, the system alerts teachers and schools, while our team also reviews sample conversations. Together, these safeguards help ensure BloomyBot provides reliable, appropriate guidance while keeping teachers informed.
Kertai: What types of insights does Bloomy provide to teachers?
Southmayd: Bloomy provides reports at the individual student, small-group, classroom, and school levels. By capturing detailed learning data, including how long students spend on activities and how quickly they master new skills, the platform gives teachers clear insight into skill-level performance, overall progress, and growth over time.
One of Bloomy’s most valuable features is its ability to identify patterns across students. The platform can group students struggling with the same concept and automatically generate instructional plans tailored to each group. As Bloomy continues to learn from more data, our goal is to move beyond tracking student progress to predicting where students are most likely to struggle next, allowing teachers to intervene before confusion becomes a larger obstacle. Ultimately, we want to give teachers the information they need to provide every student with the individualized support that was once only possible through one-on-one tutoring.
