Frizzle

Frizzle is an AI-powered platform designed to fundamentally change how math teachers assess and understand student learning. It uses a combination of computer vision and large language models (LLMs) to read and grade handwritten math work with 97% accuracy. The core value proposition is simple: students continue to work on paper, using their existing workflow, while teachers gain the real-time, granular data of a digital platform. Frizzle does not simply mark answers right or wrong; it analyzes the entire problem-solving process, identifying specific steps, multiple solution paths, and 147 named misconceptions. The system features a confidence-interval mechanism that flags uncertain grades for human review, ensuring reliability. For teachers, this means reclaiming the 10 to 15 hours per week typically spent on grading. For coaches and administrators, Frizzle provides live, standards-level formative analytics aligned with frameworks like CCSS and TEKS, enabling targeted instructional conversations and district-wide insights. The product is live in over 30 schools and districts, including a college math pilot at Vanderbilt and Arizona State University. It is FERPA and COPPA compliant, ensuring student data privacy is maintained. Frizzle is built for individual teachers, schools, and entire districts, offering a free tier for individual educators. It has already graded over 100,000 questions and is used by more than 2,400 teachers.

Step-by-Step Work Recognition

Frizzle goes far beyond simple answer checking. Its computer vision model is trained on 1.4 million pages of K-12 student work to parse each individual step of a student's handwritten solution. It can recognize print, cursive, and even scribbled handwriting. This allows the system to award partial credit for correct reasoning even if the final answer is wrong, and to pinpoint the exact moment a student's thinking went off track. This provides a level of detail that traditional grading cannot offer.

Misconception Mapping and Tracing

The platform has a built-in library of 147 named misconceptions across K-12 math, each mapped to specific educational standards. Frizzle does not just flag a wrong answer; it identifies the underlying misconception, such as a sign error or a misunderstanding of the distributive property. It also features prerequisite tracing, which can identify when a student's error in 7th grade is actually rooted in a foundational gap from 4th grade, allowing for truly targeted intervention.

Live Classroom and District Dashboards

Frizzle provides a live, updated dashboard that gives an immediate snapshot of an entire class or district. For a teacher, this shows which students are stuck, which misconceptions are spreading, and what concepts need to be retaught the next day. For a school or district, aggregated and anonymized dashboards reveal performance gaps across periods, grades, and buildings, showing which curricula are effective and where to invest resources. This creates a real-time "nervous system" for math instruction.

Confidence-Interval Flagging for Accuracy

To maintain a 97% accuracy rate, Frizzle employs a sophisticated confidence-interval system. When the AI model is not completely certain about a grade or a specific step in a student's work, it flags that paper for human review. This ensures that teachers only need to look at a small subset of papers where their judgment is most critical, rather than re-grading the entire stack. This system builds trust in the AI's output while maintaining the highest standard of accuracy.

Reclaiming Teacher Time for Instruction

A high school Algebra I teacher currently spends 10 to 15 hours each week grading stacks of paper assignments. By using Frizzle, they can simply photograph a stack of papers with their phone or run them through a copier. In about 30 seconds for the upload and roughly 8 minutes for processing, the entire class is graded. The teacher can then use the reclaimed time to plan more effective lessons, provide one-on-one support to struggling students, or collaborate with colleagues.

Targeted Intervention for At-Risk Students

A middle school math coach uses Frizzle's dashboards to identify students who are "At Risk" across the entire 7th grade. The system flags a student, Jaden K., who is at 52% mastery with a specific "Distributive" misconception. Instead of a generic conversation, the coach and teacher can now have a specific, data-driven discussion about Jaden's needs. They can review the exact step where the error occurred on his paper, allowing them to design a targeted micro-lesson to address the precise gap.

District-Wide Curriculum Analysis

A district math coordinator wants to know if a new Illustrative Math curriculum is more effective than the previous Eureka program. Frizzle provides aggregated, curriculum-agnostic data showing standards-level mastery across all schools in the district. The coordinator can see which specific CCSS standards are being mastered and which are not, broken down by school and teacher. This data enables the coordinator to make informed decisions about professional development and resource allocation, moving beyond anecdotal evidence.

Formative Assessment Without Screen Time

An elementary school is concerned about increasing students' screen time but still needs robust classroom-level data. Frizzle allows students to continue doing all their math work on paper, which is crucial for developing fine motor skills and reducing digital fatigue. Teachers can still run quick, daily formative assessments by photographing the paper exit tickets. The platform provides immediate analytics on class-wide understanding of a new concept, all without requiring a single student to log into a device.

How does Frizzle handle different handwriting styles?

Frizzle's AI model has been trained on over 1.4 million pages of real K-12 student work, which includes a vast range of handwriting styles, from neat print to messy cursive and sideways scribbles. The computer vision is designed to be robust and recognizes the intent of the work, not just perfectly formed characters. It can also understand multiple different solution paths to the same problem, ensuring students are not penalized for solving problems in their own way.

Is my students' data safe and private?

Yes, data privacy is a foundational element of Frizzle's design. The platform is fully compliant with FERPA and COPPA regulations and undergoes annual SOC 2 Type II audits. All student data is encrypted with AES-256 at rest and TLS in transit. Crucially, student work is never used to train or improve the underlying AI model, meaning your students' data remains completely yours and is not used for any other purpose.

Does Frizzle work with my specific curriculum?

Yes, Frizzle is designed to be curriculum-agnostic. It works with any math curriculum, including common ones like Eureka, Illustrative Mathematics, and Saxon. The platform aligns its analytics to major state and national standards frameworks, including CCSS, TEKS, and over 30 other state frameworks. This allows teachers to get standards-level data regardless of the specific textbook or lesson materials they are using in their classroom.

How accurate is Frizzle's grading, and what happens if it is unsure?

Frizzle grades handwritten math with a verified 97% accuracy rate. To achieve this, the system uses a confidence-interval system. When the AI is highly confident in its assessment, it grades the paper automatically. When it encounters a problem or a step where its confidence is lower, it flags that specific paper for human review. This ensures that a teacher's time is spent only on the most ambiguous cases, maintaining high accuracy without requiring a full manual re-grade.

Frizzle is free for individual teachers to get started. The platform offers a "Start free" option with no credit card required. For school and district-wide implementations, pricing is available upon request to provide tailored solutions for larger organizations.