Too Long Don't Read

Building Thai AI that understands more than words.

Meet THIRA

THIRA — Thai Handwriting Intelligence Recognition & Analysis

Thai handwriting carries more than text. It contains mathematical notation, working steps, and evidence of how a student thinks. THIRA is our research initiative to make that information more accessible to educational AI systems.

We are developing AI components that can read handwritten Thai, recognize mathematical expressions, understand worksheet structure, and support the analysis of student work. Our goal is to turn handwritten work into useful, reviewable information—not simply extract text from an image.

Our mission: Make Thai handwriting understandable to AI, so teachers can spend more time understanding their students.

What we're building

From research to the classroom

THIRA is being developed as part of the AI foundation for Learnly, our educational technology project. Learnly explores how AI can help teachers review student work, identify possible mistakes, and prepare feedback for teacher approval.

Our approach keeps the teacher in control: AI assists with reading and analysis, while uncertain findings and feedback remain subject to human review. We believe useful educational AI must be evaluated on real student work, not only clean benchmark images.

Current research focus

We are actively experimenting with vision-language models, task-specific adapters, and evaluation pipelines for Thai handwriting and mathematical notation. Our work includes document-region detection, OCR benchmarking, structured transcription, and end-to-end worksheet analysis.

We are especially interested in the gap between strong benchmark performance and reliable performance on real classroom documents. That means measuring errors carefully, testing on held-out material, and making uncertainty visible rather than overstating accuracy.

Research status: THIRA is under active development. Models and repositories published here may be experimental and should not be treated as independently validated or production-ready unless their model cards explicitly state otherwise.

Our team

We are TLDR (Too Long Don't Read), a five-person team bringing together entrepreneurship, product development, and AI engineering.

Khon Kaen Business School · Digital Entrepreneur

What you'll find here

As our research progresses, we plan to share model checkpoints, evaluation results, technical documentation, and selected research artifacts where licensing and data permissions allow. Each release will describe its intended use, limitations, and available evaluation evidence.

We welcome conversations with researchers, educators, and developers working on Thai language technology, handwriting recognition, document AI, and responsible educational applications.