We are the Frontier of
Document Processing
AI-powered Document understanding, Knowledge Retrieval, and Automation Pipeline from Seoul, Korea.
We are the Frontier of
About Us
Iruri Labs builds production-grade document intelligence for teams that must turn real-world paperwork into reliable, structured, and actionable data. We combine semantic document understanding, knowledge retrieval, and workflow automation into a single engine—designed to be accurate, scalable, and easy to integrate.
Document Parsing
Knowledge Graph RAG
Automation Pipelines
Multilingual Handwriting
Technology
Semantic Document Parsing
We treat a document not as a flat image, but as a structured space made of semantic regions. As you scroll, the 3D document behind this page splits into paragraphs, figures, tables, and form fields—then connects them as a graph.
Layout-aware region segmentation
Robust chunking & segmentation that preserves reading order, hierarchy, and visual intent.
Graph-native representation
Regions become nodes; references and context become edges—ideal for retrieval and automation.
Integration-ready output
Export clean, consistent structures for downstream pipelines and databases.
World #1 Knowledge Graph RAG
Our Retrieval-Augmented Generation stack is built on a Knowledge Graph foundation, with semantic chunking designed for rich documents. It is optimized for precision, ranking quality, and short time-to-confidence.
Precision
0.732
MRR
0.784
F1@10
0.569
Semantic chunking for long documents
Chunks align with meaning (not arbitrary token limits), improving retrieval and reducing hallucinations.
Knowledge Graph retrieval
Graph signals help connect dispersed evidence across sections, pages, and languages.
Production stability
Deterministic pipelines, monitoring hooks, and evaluation-friendly outputs.
Document Processing Pipelines & Multilingual Handwriting
Documents become valuable only when they move through a reliable engine. As you scroll, the 3D document flows through our pipeline—getting parsed, normalized, interpreted, and enriched into machine-usable meaning.
Stable pipelining for automation
Composable stages with robust error handling, retries, and consistent intermediate formats.
Context-aware multilingual handwriting OCR
Reads handwriting across languages and scripts with context-sensitive decoding.
End-to-end data products
From raw scans to searchable records, structured tables, and downstream decisions.
Our Services
Our flagship product turns any exam paper—commercial, custom-made, or legacy—into a searchable database with minimal change to your existing teaching workflow. Scan or photograph, and we handle the rest: recognition, structuring, grading, and reporting.
Universal Exam Ingestion
Recognize questions and content from virtually any layout. Convert scans and photos into a clean, queryable database.
99% Accurate Auto-Grading
High-precision scoring for objective formats and handwriting-enabled workflows, with consistency you can audit.
Actionable Feedback Reports
Detailed, report-style feedback that helps instructors maximize learning outcomes and personalize guidance.
Easy Adoption for Any Institution
Deploy without disrupting your existing class operations. Keep the same teaching style—upgrade the workflow.
Meet the Team
Our Team
Engineers from SNU. Experts in AI, NLP, and Software Engineering.
Located at Seoul
At the heart of the Asian AI community, pushing through the boundaries of AI.
Collaboration
We work together to solve hard problems.