Postmath is applying knowledge graphs to education, representing a unknown vertical AI play with core generative AI integration.
As agentic architectures emerge as the dominant build pattern, Postmath is positioned to benefit from enterprise demand for autonomous workflow solutions. The timing aligns with broader market readiness for AI systems that can execute multi-step tasks without human intervention.
Postmath offers AI-driven SaaS solutions that automate content generation and streamline mathematical processes for digital transformation.
A domain-specific stack combining math-optimized OCR, a curriculum-aware semantic/tagging system (Adaptive TAG + DSS + Math Ontology), visual-matching (LDM), and a topology-preserving geometry editor (Postgeo) that together enable automated, high-quality conversion and generation of math content into a structured problem bank.
They maintain a domain-specific ontology/taxonomy that links math problems and concepts via common tags and tree structures, enabling semantic connections, structured search, and curriculum alignment.
Emerging pattern with potential to unlock new application categories.
They appear to combine structured/semantic indices (ontology + DSS) with retrieval for content search and assembly. While not explicit about LLM generation pipelines, the presence of semantic structuring and a problem bank suggests retrieval-backed content generation or augmentation.
Emerging pattern with potential to unlock new application categories.
User assessments (OMR/review surveys) are used to measure item efficacy and automatically select high-utility items for re-learning, creating an operational feedback loop where usage and outcome data inform content curation and likely model/data updates.
Winner-take-most dynamics in categories where well-executed. Defensibility against well-funded competitors.
They curate a proprietary, industry-specific dataset (math problem bank, labeled concept links, usage metrics) and productize it (books, platform), creating domain-specific defensibility and training data advantages.
Unlocks AI applications in regulated industries where generic models fail. Creates acquisition targets for incumbents.
Postmath builds on unknown, leveraging unknown infrastructure with unknown in the stack. The technical approach emphasizes hybrid.
Not explicitly specified in the provided content; appears as founder/CEO and is associated with multiple notable achievements within the company context
Limited information on multiple founders; however, the CEO identity and the company's product portfolio indicate alignment with the AI/EdTech and math education problem space. Additional founder details would strengthen assessment.
partnership led
Target: enterprise
hybrid
AI 기반 수학 콘텐츠 제작 및 관리(문제은행, 태깅, 도형/그래프 편집, 개념 연결)
Mathematical figure similarity demands respecting geometric/semantic invariants (not just pixel similarity). A specialized image-matching model for math diagrams suggests custom embeddings or descriptor extraction tuned to topology/geometry.
Using topology/constraint models in an interactive geometry editor is relatively specialized — it lets teachers/students manipulate diagrams while automatically preserving mathematical properties, enabling robust content creation and reuse.
Postmath operates in a competitive landscape that includes Mathpix, Photomath, QANDA (Mathpresso).
Differentiation: Postmath couples math OCR with an end-to-end teacher-focused content pipeline (problem bank, adaptive TAG, Math Ontology, DSS, OMR review, and content-generation SaaS). It also includes proprietary geometry editing (Postgeo) and visual matching (LDM) that Mathpix does not offer as integrated teacher tooling.
Differentiation: Photomath focuses on learner-facing solution workflows. Postmath is enterprise/SaaS and teacher-facing: automating content generation, curating problem banks, tagging problems to curricula via Adaptive TAG and Math Ontology, and providing tools to create/organize content at scale for classrooms and publishers.
Differentiation: QANDA is primarily a student Q&A/search service. Postmath positions as a content-creation and management platform for teachers and institutions (problem bank '수학비서'), with features for curriculum alignment (Adaptive TAG, DSS), item analysis (OMR review), and diagram/geometry tooling (Postgeo, LDM).
End-to-end, math-first ingestion pipeline: they combine an OCR tuned for printed+handwritten text with downstream semantic structuring and tagging rather than treating OCR as an isolated component. The emphasis is on converting messy problem pages (typed, handwritten, doodled) into structured problem objects that include text, symbols, diagrams and curriculum metadata.
OCR claims go beyond standard printed-text OCR — likely doing stroke/ink segmentation + symbol recognition + layout parsing so it can recover mixed printed/math handwritten problems with scribbles. That implies specialized training data, multi-stage models (segmentation → stroke-vectorization → math-symbol recognition → semantic parsing) or multimodal transformers with layout awareness.
Adaptive TAG = hybrid hierarchical taxonomy + tag graph. Instead of flat tagging, they combine tags with a tree structure to automatically analyze and reorganize problems across multiple curricula. This is a curriculum-mapping layer that normalizes the same problem into different educational standards/levels.
DSS (Dynamic Semantic Structuring) suggests automated extraction of multi-granular textbook/reference information (definitions, theorems, worked examples, problem contexts) and dynamic linking to problem instances to maximize searchability and reuse. This is essentially building a rich, layout-aware semantic index of learning materials.
Topology-based geometry editor (PostGeo) that preserves geometric properties on drag & drop. That indicates a constraint-based geometry engine (symbolic constraints + numerical solvers) rather than simple vector drawing — enabling parametric manipulation of figures while keeping invariants (parallelism, congruence, circle tangency, etc.).
Postmath's execution will test whether knowledge graphs can deliver sustainable competitive advantage in education. A successful outcome would validate the vertical AI thesis and likely trigger increased investment in similar plays. Incumbents in education should monitor closely for early signs of customer adoption.
“‘AI 기반 수학 콘텐츠 제작 플랫폼’”
“AI기술”
“LDM 수학적으로 유사한 그림을 자동 매칭하는 기술로, 문제 속 도형이나 그래프를 분석해 의미적으로 유사한 그림을 찾아주는 AI기술입니다.”
“Topology-preserving geometry editor (포스트지오): a drag-and-drop, topology-based shape editor that maintains geometric properties while enabling easy editing — a domain-specific UI + geometry/constraint engine.”
“LDM for mathematically similar image matching: an image-matching approach specialized for math diagrams/graphs that finds semantically/mathematically similar figures (likely a learned embedding/matching model tuned for mathematical invariants).”
“Adaptive TAG combining tags + tree structure for automatic curriculum-aware problem organization — hybrid of flat tagging and hierarchical taxonomy tailored to multiple curricula.”
