AIQALAM

Inspiration

Teachers in our target schools can spend around three hours after lessons reviewing handwritten work, finding repeated language mistakes, and preparing individual feedback. This work is essential, but it is slow, difficult to scale, and especially challenging for Kazakh-language handwriting, which remains underrepresented in global AI systems.

We built AIQALAM to return that time to teachers without removing their professional judgment.

What it does

AIQALAM turns handwritten notebook review into a teacher-controlled digital workflow. A teacher uploads a notebook page, AIQALAM's proprietary OCR reads and structures the Kazakh or Russian handwriting, and Gemini analyzes the recognized text in context. Gemini identifies likely spelling or grammar issues and drafts a clear explanation. The teacher can inspect the original work, correct, reject, or approve every recommendation.

AIQALAM is not an autonomous grader. The teacher always makes the final educational decision.

How we built it

The current competition version was started after 19 May 2026. We combined two complementary AI layers:

1. AIQALAM's proprietary OCR recognizes and structures local-language handwriting.
2. Gemini provides contextual language analysis, identifies likely errors, and drafts teacher-facing explanations.
3. AIQALAM validates the structured response before it is shown.
4. The teacher reviews the recommendation and makes the final decision.

A privacy-minimized production audit on 5 August 2026 recorded 37 Gemini events: 35 completed and two failed. It recorded 83,943 prompt tokens, 9,734 output tokens, and a 94.59% successful execution rate. The current verified production model is gemini-3.1-pro-preview.

Traction and accomplishments

AIQALAM is running a free pilot across four schools, two classes per school: eight classes and approximately 90 children.

Product-development evidence includes:

• 6,000 real notebook pages processed;
• 125,000 annotated Kazakh words;
• 94% measured OCR accuracy;
• four pilot schools, eight classes, and approximately 90 participating children.

The pilot is free by design. Our current goal is to validate classroom value, teacher trust, and workflow quality before commercial rollout.

Challenges we ran into

The hardest challenges are the diversity of real handwriting, limited high-quality Kazakh-language data, false positives, and the need to protect student privacy. Accuracy alone is not enough in education. Teachers need to see the original work, understand why an issue was flagged, and remain able to correct or reject every recommendation.

We therefore treat Gemini as a contextual reasoning and explanation layer, while AIQALAM's OCR supplies the local-language specialization and the teacher-controlled workflow provides accountability.

What we learned

Our pilot reinforced that useful education AI must earn teacher trust. The best system is not the one that makes the most decisions automatically; it is the one that removes repetitive work while making every recommendation reviewable.

Humans and AI have deliberately separate responsibilities. AI handles repetitive reading support, contextual error analysis, and first-draft explanations. Teachers provide curriculum context, professional judgment, corrections, and the final decision.

What's next for AIQALAM

The commercial model combines annual school licensing, paid institutional pilots, and integrations with education platforms. We will expand from Kazakhstan to selected CIS markets with similar language and handwriting challenges, followed by selected European markets.

AIQALAM can reduce unpaid after-hours workload for teachers, increase timely feedback for students, and create demand for local-language data annotators, education methodologists, onboarding specialists, and integration partners. Most importantly, it builds technology and data capacity for Kazakh, a language still underserved by global education AI.

Built With

• artificial-intelligence
• edtech
• gemini-api
• google-cloud
• ocr
• web