Massive documents with tables
Corporate policies and portfolio reports of 500+ pages, with repeating tables, consolidated into one JSON in minutes. In production today.
orbit vision
/
Extraction · Documents
Any document, ready for your AI.
orbit vision turns any PDF or spreadsheet, digital or scanned, even hundreds of pages with tables, into text with the layout intact or structured JSON, through a single API call. In production with active clients.
orbit vision
extraction · documents
01 · Context
The problem we are solving.
Policies, endorsements, invoices, and portfolio reports arrive as PDF or Excel: formats made for people, not systems. The PDF that will not copy gets typed by hand, tables break with generic tools, and per-page services cost more than manual capture once a document passes a hundred pages. orbit vision turns that input into text a language model reads right away.
02 · What it is
orbit vision in one sentence
A REST API that takes a PDF or spreadsheet, figures out on its own whether it is digital, scanned, or massive with tables, and returns its content as text with the original layout preserved or as structured JSON.
03 · How it works
Four steps. No magic.
Not a magic button. A clear, governed and auditable process that connects with what you already have.
01
The document comes in
Your system sends a POST /extract with the file URL or its base64 content. Page ranges are supported.
02
It figures out what it is
Identifies the file type and checks, page by page, whether the text is native or scanned.
03
It adapts to the document
Reads digital text directly, interprets scanned pages with vision, and consolidates repeating tables. A mixed PDF comes back whole.
04
It comes out ready to use
Returns JSON with the content and its layout intact: columns, tables, and page breaks, plus processing metadata.
Example · schema configured for policies live
Example · schema configured for contracts live
04 · What it does
Where it goes to work.
Two of these cases already run in production with active clients; the rest apply the same pipeline to other documents.
Images inside PDFs
Logos, charts, and photos embedded in digital PDFs, read as part of the content. In production today.
Policies and quotes
Insured party, terms, coverages, premiums, and exclusions extracted to feed comparison or management.
Invoices and purchase orders
Folio, vendor, line items, amounts, and due dates from digital or scanned documents.
Contract review
Clauses, parties, dates, and obligations identified, with the output delivered straight to a model.
Physical files
Digitized paper turned into queryable, structured content.
05 · Benefits
What changes when you have it.
We do not promise transformation. We promise measurable differences from month one.
Request demo1
One step to your AI
The text it returns goes straight into the model prompt, with no parsing or cleanup layers in between.
2
Cost does not grow per page
Digital and massive table documents never call an external model, so the price does not scale with page count.
3
Tables arrive whole
Columns, alignment, and table structure preserved: the model knows which value belongs to which field.
4
Mixed PDFs handle themselves
Digital and scanned pages in the same file, each handled the right way, without splitting the file first.
5
It plugs in where you already work
Reads straight from your cloud storage and runs as an HTTP step in your code, your orchestrator, or an agent.
06 · Industries
For sectors where errors are expensive.
With a real case in insurance today; the same document pipeline applies wherever paper slows the operation.
01 / Insurance brokers
500-page portfolio reports, consolidated in minutes
Coverage tables that used to be typed by hand, turned into one JSON. In production today.
02 / Insurers and reinsurers
Policies, endorsements, and bordereaux into the same pipeline
The brokers' same document case, applied to underwriting and claims operations.
03 / Legal and document operations
Scanned files turned queryable
Contracts and digitized paper archives, ready for model-assisted review.
07 · Applications
Concrete ways to apply it.
Application · 01
Bordereaux and portfolio consolidation
Documents hundreds of pages long with the same repeating table, turned into one JSON.
APPLICATIONApplication · 02
Policy and endorsement intake to the core
The document goes in and the data comes out structured, with no manual capture in between.
APPLICATIONApplication · 03
Scanned archive rescue
Digitized physical files that no system can query today, turned into processable content.
APPLICATIONApplication · 04
Reading layer for agents and RAG
The document ingestion step of an AI pipeline, callable by the agent as a tool.
APPLICATION08 · Before and after
What changes, side by side.
An honest comparison between how your team operates today and with orbit vision.
Before / Without orbit vision
Today's day-to-day
- ✕ The PDF will not copy, so someone types it by hand
- ✕ The scan of the paper archive is a dead file
- ✕ Tables break with generic tools
- ✕ Per-page cost ends up above manual capture
- ✕ Text with no spatial context that demands parsing and cleanup
After / With orbit vision
The day-to-day with AI
- ✓ Text with the layout intact, straight into the prompt
- ✓ Whole tables: every value in its field
- ✓ A mixed PDF comes back whole, no pre-splitting
- ✓ No per-page cost on digital and massive documents
- ✓ One HTTP call from your code, orchestrator, or agent
09 · Metrics
The numbers that matter.
Technical references for the product, each with its source. The formal benchmark is underway: no figure gets published without saying where it comes from.
50 págs
In under 2 seconds · digital document, PRD reference
1
Integration step to the first model call · verifiable in the API
~100%
Accuracy on digital documents: the file's native text is read
95-97%
Accuracy on scanned documents, depending on image quality
10 · Impact · No brainer
Paper stopped being a barrier.
What used to live in filing cabinets, bad scans, and shared folders is now clean data in your critical systems, ready for decisions.
Run the discovery with rocky<2 s
50 digital pages
1
Step to your model
~100%
Accuracy on digital
95-97%
On scanned
11 · Security and compliance
Built for regulated sectors.
If your industry does not forgive errors or leaks, this is the floor, not the ceiling.
Encryption
In transit and at rest
TLS 1.2+ and storage encryption across all environments.
Digital
Never leave to external models
Digital and massive table documents are processed without calling any third-party model.
Scanned
Third-party vision, disclosed
Scanned pages are interpreted with an external vision model: we tell you in the very first conversation.
Deployment
SaaS with scale-to-zero
Runs on Google Cloud Run; massive documents use a dedicated instance.
12 · Contact
Let us get orbit vision running in your company.
30-minute demo with your real data. If it fits, we define a measurable pilot. If not, we tell you which orbit product does fit.