Grade notebooks
in 15 minutes

Instead of 3 hours with a red pen. AI built for schools in Kazakhstan — reads handwriting, flags errors, prepares the grade.

15 minInstead of 3h/day

<5sPer page

125KTraining samples

94%Character accuracy

Q QALAM

◉ Dashboardnew

◎ Classes4

◎ Assignments3

◎ Grading12

◎ Analytics

◎ Archive247

АН

Aliya N.

Teacher · 5-A

QALAM OCR2 сек

Analyzing

94%

АС

Aigerim S. 5-A · Kazakh

Grading

23lines

3errors

94%accuracy

Found by AI

Grammarp. 2

Diacritic ө→оp. 5

Spellingp. 8

Approve · grade 4

→ Save and send to student

Teachers in Kazakhstan are drowning in notebooks

3 hours a day, red pen, 30 students × 5 subjects. No one built OCR for Kazakh handwriting — until us.

3 hr per day a teacher spends grading notebooks after class

8,460 checks per year a single teacher grades by hand (30 × 2 × 141 school days)

0 off-the-shelf OCR solutions on the market for Kazakh handwriting

! Today

Manual grading eats the evening

3 hours a day the teacher grades by hand — after class, at home
Kazakh handwriting (ә, ғ, қ, ң, ө, ұ, ү, і) is not handled by Google OCR or any Western service
The ChatGPT epidemic — students submit AI-generated texts; you can't tell by eye
No analytics — recurring class errors get lost; reports are written in Word
Gradescope and similar tools are built for Latin script — nothing exists for Kazakh + Russian Cyrillic

✓ With QALAM

15 minutes instead of 3 hours

Photo → graded in 5 seconds — upload a page, get a marked-up result
Our own OCR trained on 125K Kazakh handwriting samples — built for ә, ғ, қ, ң, continuously learning on school notebooks. Patent application filed in Kazakhstan
AI-powered ChatGPT detector — compares style with the student's prior work
Auto-reports and analytics — per-student progress, common class errors, parent-meeting summaries ready to share
The first and only in Kazakhstan — 2 languages (Kazakh + Russian), our own model trained on a Kazakh dataset from local students

Why no one solved this before us

Western OCR is built for Latin script

Google Cloud Vision, AWS Textract, and Tesseract were trained on English typography. They misread Kazakh handwriting — diacritics drop out.

No open dataset of Kazakh school notebooks

There is no labeled handwritten dataset from Kazakh schoolchildren in any open source. We had to build it from scratch — that's our moat.

Gradescope and similar tools target printed text

Western EdTech platforms grade printed text, PDFs, and multiple-choice tests. They were never built for Kazakh handwriting.

Diacritics break the models

Letters like ә, ғ, қ, ң, ө, ұ, ү, і rarely appear in mainstream datasets. Without specialized training, models confuse them with similar Russian letters.

Features

Everything you need to grade notebooks

125K labeled words · 8 Kazakh letters

Kazakh handwriting recognition

Our own neural network trained on 125,000 Kazakh handwritten words with diacritics, continuously fine-tuned on real school notebooks. Supports every special letter of the Kazakh alphabet: ә, ғ, қ, ң, ө, ұ, ү, і. Handles messy handwriting.

94% accuracy · <5 sec

Automatic error detection

Spelling, grammar, diacritics, punctuation. Highlighted directly on the photo.

GPT detector

AI cheating detector

Detects ChatGPT by writing style. Cross-checks with the student's prior work.

Annotations on the photo

Red pen 2.0

Circles, arrows, comments on top of the photo. Like on paper — but archived.

Reports in one click

Progress analytics

Per-student trends, recurring class errors, parent-meeting summaries.

30 notebooks at once

Batch grading for a whole class

Upload photos for the whole class as one batch — AI processes them sequentially and produces a summary.

How to onboard a school

Four stages — from a free pilot to a full school-wide rollout.

Pilot in one class

1 week, free. One teacher uploads notebooks, evaluates the results — we deliver a report.

Teacher training

2-3 hour workshop for the whole school. Video guides, instructions, Q&A.

School-wide launch

We connect every class and configure subjects and grading rubrics for the school's methodology.

Monitoring and support

Regular reports, feedback collection, fine-tuning the model to the school's methodology.

Model parameters

OCR accuracy, training dataset size, grading time benchmarks.

OCR performance

Kazakh handwriting

94%character accuracy

<5sper page

2languages: KK + RU

Teacher's time

Grading a class of 30 notebooks

180 minBy hand, red pen

→

15 minWith QALAM

12× faster than manual grading.

Training dataset

Proprietary Kazakh corpus

125Klabeled words

2–11school grades

8Kazakh letters: ә ғ қ ң ө ұ ү і

Roadmap

Where we are, what we're building, what we're planning — no fluff.

Q1 2026

Q2 2026now

Q3 2026

Q4 2026

Q1 2027

Q2 2027

✓ Shipped

Kazakh handwriting OCR · 94%

Web platform · upload, annotate

8 Kazakh special letters

Servers in KZ · TLS 1.3 / AES-256

◐ In progress

Fine-tuning OCR for full lines

Error highlights on the photo

Telegram bot for requests

Full roles (teacher / vice / principal)

Launch of school pilots

→ Planned

School hierarchy (classes, students)

Per-student & per-class analytics

Mobile PWA version

AI plagiarism detector

Shipped In progress Planned

Security and compliance

Student data is protected to Kazakhstan and international standards.

Servers in Kazakhstan

All student data is stored on servers in Kazakhstan. Compliant with Law No. 94-V "On Personal Data and Its Protection".

TLS 1.3 · AES-256

TLS 1.3 in transit, AES-256 at rest. Access only via school accounts.

School-only access

Data is never shared with third parties. A parent may request deletion of their child's work at any time.

About us

QALAM is built by a solo founder from Kazakhstan. One developer building what Big Tech didn't: AI tools for schools — Kazakh handwriting OCR, automated grading, analytics.

Ilias Kabden

Founder · CTO

Full QALAM stack end-to-end: training the CRNN model on Kazakh handwriting (CER 5.66%, 125K samples, RTX 3090), our own production OCR pipeline, deployment, product and partnerships with Kazakh schools.

Google AI Certified ML / OCR Full-stack

Certificates and education

Verified credentials of the founder — checkable on Coursera.

Google · Coursera

Google AI Professional Certificate

7 courses: AI Fundamentals, Brainstorming & Planning, Research & Insights, Writing, Content Creation, Data Analysis, App Building. May 2026.

Verify →

Why not ChatGPT, Claude, or Google?

General-purpose LLMs and OCR were never trained on Kazakh handwriting. QALAM is the only model purpose-built for Kazakh schools.

QALAM

ChatGPT

Claude

Google Vision

Kazakh handwriting

✓

—

Diacritics ә, ғ, қ, ң, ө, ұ, ү, і

✓

—

Reads notebook photos

✓

Servers and data in Kazakhstan

✓

—

School-wide license (not per-user)

✓

—

Work archive + class analytics

✓

—

AI ChatGPT detector for student work

✓

—

Grade notebooksin 15 minutes