May 22, 2026

Why I Built GradeSnap: A 6th Grade Teacher's Year of AI, Code, and a 32-Point RISE Jump

It was the end of last spring's 4th term, and my dining room table had disappeared under a stack of paper roughly two feet tall. Twenty-five different assignments, all jumbled together — math quizzes on top of social studies tests on top of ELA writing samples on top of science worksheets. I teach 6th grade, which means I teach every subject, which means at the end of a grading window, all of it lands on me at the same time.

I needed to sort the pile by subject. Then by assignment. Then dig out the answer key or the rubric for each one. Then grade. Then record each grade in my gradebook under the correct Utah Core Standard. Then re-enter every single score into Skyward, our district's grading system. And I needed to move fast — the grading window was closing, and my students were still waiting to find out which assignments were missing. Assignments they could still turn in, if they only knew.

I sat down in front of the stack, looked at it for a long minute, and thought: what if I could just run every one of these through a scanner — and the scanner sorted each paper into the right class, matched it to the right student, found the right answer key or rubric, graded it, recorded the score under the right standard, pushed it to Skyward — all without me touching another piece of paper? And then I could throw every last sheet away. Forever.

That was the moment GradeSnap started.

I'm a 6th grade teacher at Alpine School District in Utah. I'd been teaching long enough to know two things were true at once: the grading I was doing was genuinely valuable for my students, and the way I was doing it — late nights, weekends, double-entry into two different systems, paper piles I couldn't see the bottom of — wasn't sustainable.

This is what I built instead, and what happened when I used it in my own classroom.

The numbers that made me stop and stare

Before I tell you about the code, the late nights, or the year of learning to build software from a dining room table — let me show you the result.

These are end-of-year Utah RISE proficiency scores for my 6th grade students, comparing their 5th grade year to their 6th grade year with me:

• ELA: 26% proficient → 58% proficient (+32 percentage points)
• Math: 32% proficient → 56% proficient (+24 percentage points)
• Science: 46% proficient → 69% proficient (+23 percentage points)

The first time I saw those numbers, I assumed I was misreading them. Pre-pandemic cohort analyses generally show about one percentage point of proficiency growth per year as a healthy baseline. State-level "turnaround" stories — Mississippi's 4th-grade math gains, Alabama's ELA jump — happened over multiple years across entire states and were considered exceptional.

But raw growth is only half the story. The other half is what the rest of the state was doing during the same year.

In 2025, my students scored below state, district, and school averages in every subject — 18 points below the state in ELA, 12 points below in Math, 4 points below in Science. By the end of 2026, the same kids scored above state, district, and school averages in every subject:

The Math result is the one I keep coming back to. During the same school year my students gained 24 points in Math, statewide 6th grade Math proficiency fell by 7 points. The state was sliding backward in Math. My class was sprinting forward. Relative to the state trend, that's a 31-point swing in a single subject, in a single year.

I want to be careful here. One classroom is one classroom. Test scores swing more in small samples than in large ones. I'm not claiming a controlled study. I'm telling you what happened in my room, and then I'll tell you what I think the mechanism was. You can decide what to make of it.

I also want to be honest about something else: GradeSnap is almost certainly not the only reason those scores went up. Too many variables move at once in a single classroom to ever claim a single cause. The kids matured another year. Test difficulty varies year to year. I'm a teacher who's constantly trying to get better — some of what happened this year is the cumulative effect of choices I've been refining for years. This cohort showed up willing to try things. Their families backed them. Their previous teachers built foundations I got to build on.

What I can say is that GradeSnap is the one variable I can point at directly — the thing that changed in my workflow this year that wasn't there the year before. I don't know exactly how much of the jump it accounts for. What I do know is what changed about how I teach. The rest of this post is my best honest attempt to describe it.

What I was actually doing before

For most of my career, the grading workflow looked something like this. Collect a stack of paper. Carry it home. Spread it across the table at 8 PM. Read each piece, score it against a rubric, write feedback on the page, enter the score into the gradebook, repeat 29 times. Hand them back to students three or four days later, by which point most of them had emotionally moved on from the assignment.

That last part is the real problem. Not the hours of grading — I had made peace with the hours. The problem was the delay. Educational research has been screaming for decades that feedback is most useful when it's immediate, specific, and actionable. By the time I returned a writing piece on Thursday, my students were thinking about Friday's science test. The feedback was correct. It was thorough. It was just too late to matter.

I had tried every workaround. Rubric stamps. Peer editing. Shorter assignments. Faster turnaround at the cost of depth. None of it solved the underlying problem: there was one of me, twenty-nine of them, and a 24-hour day.

In mid-July 2025, I started sketching out a different idea on the back of a planning document.

The spark

The idea was simple to say and absurd to attempt: what if a teacher could photograph a stack of handwritten assignments, and an AI could read them, score them against a rubric I designed, align them to state standards, and put the results in a gradebook — all in minutes?

The pieces existed. OCR could read handwriting. Large language models could grade against a rubric if you wrote the prompt carefully. Cloud storage could hold the artifacts. State standards could be loaded into a database. None of this was novel technology. What didn't exist was a tool that put all of those pieces together specifically for the K–12 teacher with a paper assignment in her hand on Sunday night.

Before I built anything, I went looking. I spent weeks searching for the tool that already did this. There were a few that came close — tools built for higher-ed instructors, photo-grading apps for math, AI essay scorers for older students — but none of them did the specific thing I needed. None of them could take a mixed stack of K–12 paper, sort it by class and student, recognize handwriting reliably, grade against the rubrics I'd written for my own classroom, align scores to Utah Core Standards, and put the results where I could actually use them. The pieces existed in the world. The right combination, for the teacher I was, didn't.

So I decided to build it.

I knew exactly what teachers needed because I was the teacher who needed it. I had two problems, though.

First, I had never written a line of code in my life.

Second, I was teaching full time.

Learning to build from nothing

I opened a chat with Claude in mid-July and asked, in essentially these words, "Can you help me figure out if my idea is even possible, and what it would take to build it?"

That conversation became a year-long apprenticeship.

I didn't take a bootcamp. I didn't enroll in a certificate. I tried to find free evening coding classes in Utah County and there essentially weren't any. So I did what teachers do — I figured it out, one piece at a time, with a patient explainer at my elbow.

I learned what a dependency was by asking what was in my requirements.txt file and getting an answer that started with "dependencies are like ingredients in a recipe." I learned what Docker was by breaking it and getting walked through the fix. I learned what a JWT token was the third time my login broke. I learned what CORS was by spending an entire weekend on Google OAuth redirect URIs and slowly understanding why my browser kept refusing to talk to my own backend.

What started as a Google Sheets gradebook concept turned, slowly, into a real piece of software. By September I had a working local prototype. By October, my first production deployment was running on AWS. By November, I had solved the batch upload problem — teachers can now upload thirty assignments at once, and the system sorts them, identifies which student wrote which paper, and grades each one against the same rubric. By December, I had loaded more than 800 real Utah and Idaho educational standards into a database and built a system that could align student work to specific standards automatically.

Through all of it, I kept teaching. And parenting. I'm a mom of six — four still at home and two in college — and my kids are deeply involved in their own very full lives, which means I built GradeSnap in the cracks between everything else. Mornings were sixth graders. Evenings were code, usually with the TV on in the background and one of my kids wandering through every fifteen minutes to ask what was for dinner. At some point they started teasing me that I spent too much time with my "boyfriend" — meaning the AI on my laptop, the one I was learning to code with. They even gave him a name: Arlo. "How's Arlo doing?" they'd ask when I came up for air. They weren't entirely wrong.

There were days when I'd debug a database issue between bell rings. There were weeks where I felt like I was failing at all three jobs at once. There were also nights — and these are the ones I remember — where I'd ship a feature, then the next morning use that exact feature to grade a stack of student work in fifteen minutes instead of three hours.

That part still doesn't feel real, and I built it.

What changed in my classroom

GradeSnap didn't change what I taught. It changed how fast I could close the loop between teaching and student feedback, and that small mechanical change cascaded into something much bigger.

Here is what I actually did differently this year:

I assigned more writing. When grading no longer cost me my Sundays, I stopped flinching at the thought of another essay. My students wrote more. They got faster, more specific feedback. They could see their own growth across the year.

I returned work the same day or the next. Feedback landed while the assignment was still alive in their heads. Students could read a comment, look at the original prompt, and try again before they had forgotten why they wrote what they wrote.

I taught from the data instead of from a feeling. GradeSnap was scoring against the same standards I was teaching to, which meant I could open a dashboard on Monday morning and see exactly which standards my students were struggling with. Reteaching stopped being a guess. If 20 of 29 students missed a question tied to ES.6.W.1c, I reteaught that standard the next day.

I gave more low-stakes practice. Quick formatives that I would never have assigned before because the grading burden didn't justify the data — those became my best diagnostic tool. A five-minute exit ticket on Tuesday told me what to reteach on Wednesday.

None of those moves are new in education. Faster feedback loops, data-driven reteaching, more deliberate practice — these are well-established principles. What was new was that I could actually do them, consistently, without burning out.

I think that's the mechanism behind the RISE jump. Not magic, not even particularly novel pedagogy. Just enough time recovered to teach the way I had always wanted to teach.

What I would tell another teacher

A few honest things from a year of doing this:

Your time is the bottleneck, not your skill. I have not become a better writer of rubrics or a better explainer of comma rules than I was last September. I have become a teacher with more hours in my week. That alone moved scores.

Same-day feedback matters more than perfect feedback. A student who gets a B+ rubric back on Tuesday will revise. A student who gets an A+ rubric back on Friday will file it.

Be skeptical of single-classroom claims, including mine. I am one teacher in one district with one cohort. Single-classroom data is noisy. What I can say with confidence is that the workflow changed, and I have the receipts (assignment counts, turnaround times, the actual scores). What caused what, I can hypothesize but not prove. That's true of every classroom intervention, including the ones that don't come with a website.

The technology should disappear. When I'm grading well, I don't think about GradeSnap. I think about the student who picked Bhutan for her country report because no one else would, and the student whose summary of "The Killer Wave" finally clicked. The tool is supposed to get out of the way of the teaching. If it ever stops doing that, I'll know I'm building it wrong.

Why I'm building this for everyone

For a while, GradeSnap was just for me. It was the tool I needed, and I happened to be the only user.

Sometime around February, I had two separate teachers ask if they could try it. Then a third. Then a teacher in Idaho — Kimberli — used it in her own classroom, which is part of why I spent a chunk of this spring making sure GradeSnap was fully compliant with Idaho's new student data law, SB 1227, which takes effect July 1, 2026.

That's when I realized what I had actually built. Not a side project. An answer to a problem every K–12 teacher I know is wrestling with right now: the gap between the kind of feedback our students deserve and the number of hours we actually have to give it.

I'm a solo developer. GradeSnap is built and maintained by one person — a teacher who learned to code with the help of AI, deployed to AWS, integrated with Google Classroom, and is now opening it up to teachers everywhere. I built it on top of OpenAI's GPT-4o and Anthropic's Claude for grading, Google Document AI for handwriting recognition, and a small mountain of careful work to make sure student data is handled the way I would want my own children's data handled.

If you teach K–12 and you've ever stared at a stack of papers on a Sunday night and thought there has to be a better way — there is. I built it because I needed it. I'm sharing it because you do too.

A note on what's next

I'm soft-launching GradeSnap to a small group of teachers this summer, with broader availability rolling out over the following months. I'd love to hear from you if you're a teacher who wants to try it, a district leader thinking about AI tools, or another teacher-builder out there who started with an idea and a kitchen table.

You can reach me at gradesnap.org, or just reply to this post. I read every email.

Thank you for reading this far. Now go give yourself a Sunday back.

— Amber Hawkes 6th Grade Teacher · Founder, GradeSnap Alpine School District, Utah