Tables
Preserve spanning headers, merged cells, footnotes, and captions as real table structure.
AlphaAccuracy-first scientific parsingAccuracy-first parsing for complex scientific PDFsSee what it preserves
Scientific document parsing
State-of-the-art parsing for scientific documents.
Turn complex scientific PDFs into structured data without losing tables, equations, captions, reading order, or the relationships that carry meaning.
01
Whole-document understanding
02
Complex-layout fidelity
03
Structured scientific data
Complex scientific PDFComplete document structure
Built for scientific content teams
Publishers
Repositories
Life-science R&D
01 / 04Accuracy
Scientific documents are different
A paper is more than the text on its pages. Its meaning lives in structure, notation, and relationships.
Equations
Keep notation, indices, equation boundaries, and numbering intact from page to output.
Reading order
Reconstruct the intended sequence across columns, floating elements, and page boundaries.
Figures & captions
Keep figures, captions, labels, and surrounding discussion connected throughout the paper.
Parser
/ understand
Reconstruct the complete paper instead of flattening one page at a time.
/ read
/ structure
/ connect
/ output
Source pagep.12
Sections
Tables
Equations
Figures + captions
One connected representation of the complete paper
Capabilities
Layout-first analysis
Understand geometry and element ownership before flattening anything into text.
Whole-document assembly
Build a single representation across every page instead of parsing pages in isolation.
Scientific element structure
Represent tables, equations, figures, captions, and references as first-class objects.
Relationship preservation
Keep reading order and connections between elements intact across the entire paper.
Optimization
Accuracy first
Fidelity before throughput
Coverage
Whole document
Cross-page structure preserved
Representation
Connected structure
Scientific elements remain related
02 / 04System
Accuracy-first architecture
Built for the layouts that break general-purpose parsers.
Amanuensis combines complementary OCR and document-understanding models, then uses automated reasoning to assemble one coherent representation of the complete paper.
- 01
Ingest
Send a scientific PDF through the API or a document processing workflow.
- 02
Analyze
Complementary OCR, layout, and vision models read every part of the paper.
- 03
Reconstruct
Our assembly system rebuilds structure, reading order, and cross-page relationships.
- 04
Verify
Automated reasoning checks the document for structural and semantic consistency.
- 05
Deliver
Receive one coherent, machine-readable representation of the complete document.
Multi-model understanding
Combine OCR, layout, and vision systems so no single model's blind spots define the result.
OCR
Layout
Vision
Reading order across pages
Reconstruct sequence across headings, columns, floating elements, and page boundaries.
01Heading
02Paragraph
03Table + caption
04Following paragraph
True table structure
Preserve headers, cells, spans, footnotes, and captions as structured data instead of flattened text.
PDF table
Table structure
Scientific relationships
Keep figures with captions, equations with numbering, and citations with references.
TableCaption
FigureCaption
EquationNumber
CitationReference
03 / 04Difference
Accuracy, not just throughput
Built for fidelity before volume.
General-purpose parsers are often optimized to make large collections searchable quickly. Amanuensis is built for workflows where a misplaced equation or malformed table changes the data itself.
The complete document is the unit of understanding
Scientific elements remain first-class structure
Accuracy is prioritized before latency and volume
| Decision point | General-purpose parser | Amanuensis |
|---|---|---|
| Primary optimization | Fast, broad document ingestion | Maximum scientific fidelity |
| Document model | Pages, text blocks, and generic markdown | One connected model of the complete paper |
| Tables & equations | Often flattened, simplified, or detached | Preserved as structured scientific elements |
| Reading order | Inferred locally from individual pages | Reconstructed across columns and pages |
| Downstream result | Extracted text that still needs repair | Structured document data ready for pipelines |
04 / 04Use cases
Scientific content infrastructure
One parser for the scientific document lifecycle.
From publishing and repositories to life-science research, Amanuensis turns difficult papers into dependable structured inputs for the systems built on top of them.
Publishers
Turn submissions and backfiles into high-fidelity inputs for production, enrichment, and discovery.
Repositories
Make heterogeneous collections machine-readable without flattening the structure of each paper.
Life-science R&D
Transform literature and scientific reports into dependable data for search, knowledge systems, and analysis.
A dependable foundation
Scientific-document intelligence starts with trustworthy parsing.
Once the paper is structured correctly, teams can power search, extraction, indexing, summarization, and future authoring workflows from a dependable foundation.
Complex-layout support
Document-level structure
Schema-ready output
Pipeline-ready data
Request a pilot
Test Amanuensis on the papers other parsers struggle with.
Bring representative PDFs and the structure your systems need. We will define the pilot around accuracy on tables, equations, captions, reading order, and whole-document completeness.