Use cases

What teams build on the pipeline.

Four stages. One backend. Open underneath. Each use case below maps to one or more pipeline stages — Acquire, Extract, Enrich, Embed & Retrieve. Pick the stages you need; the rest stay out of your way.

document
{
"title": "…",
"sections": 3,
"blocks": 24
}
01

RAG Pipeline Ingestion

Turn a pile of PDFs, Office docs, and HTML into clean, chunked, embedded data for your vector database — in one call.

ExtractEmbed & Retrieve
RAGChunkingEmbeddings
parser.py
def extract(doc):
return parse(doc)
class Reader:
# 306 formats
{
"functions": 12,
"classes": 3
}
02

Document-Reading Agents

Give your AI agents one tool to read any document — 100+ formats, structured output, every framework.

ExtractEnrich
AgentsStructured JSONLayout
INVOICE#INV-0042
Acme Supply Co.
Brackets ×40$1,200.00
Fasteners$1,280.00
Total due$2,480.00
{
"tables": 1,
"rows": 8,
"columns": 4
}
03

Replace Legacy IDP

Swap brittle, template-based processing for one API that returns schema-mapped JSON — no templates to maintain.

ExtractEnrich
IDPTablesOCRStructured JSON
combined scan
Invoice
Contract
Cover letter
{
"ocr": true,
"fields": 9
}
04

Bulk Archive Digitization

Structure millions of legacy and scanned documents for search and ML — built for batch on a Rust-native core.

Extract
MigrationOCRBatch
parser.py
def extract(doc):
return parse(doc)
class Reader:
# 306 formats
{
"functions": 12,
"classes": 3
}
05

Syntax-aware Code Chunking

Turn source code in 306 languages into structure-aware chunks — ready to embed for code search, assistants, and RAG.

Extract
CodeTree-sitterChunking
https://example.com/blog
{
"title": "…",
"sections": 3,
"blocks": 24
}
06

Web Content Ingestion

Crawl the hard web — JavaScript-heavy sites and pages behind bot protection — into clean Markdown and structured content.

AcquireExtract
CrawlJS renderingCrawl strategiesStructured JSON
document
{
"title": "…",
"sections": 3,
"blocks": 24
}
07

Semantic Search over Documents

Build semantic search over your documents with built-in chunking and ultra-fast embeddings — no separate pipeline.

ExtractEmbed & Retrieve
SearchEmbeddingsChunking
in-memory · nothing stored
{
"ocr": true,
"fields": 9
}
08

Regulated & Self-Hosted Extraction

Run the same extraction engine self-hosted or air-gapped, for regulated data that can't leave your environment.

Extract
Self-hostedOCRCompliance
INVOICE#INV-0042
Acme Supply Co.
Brackets ×40$1,200.00
Fasteners$1,280.00
Total due$2,480.00
{
"doc_type": "invoice",
"fields": 12
}
09

Pre-Built Document Extraction

Name the document type — invoice, receipt, W-2, contract, ID — and get clean, typed fields back. No templates.

Extract
IDPFieldsSchemas
onboarding form
Full nameJordan Lee
Start date2026-06-01
DepartmentFinance
{
"form_fields": 9,
"source": "AcroForm"
}
10

Form-Field Extraction

Read the answers straight out of fillable PDFs — exact field names and values, not OCR guesses.

Extract
IDPFormsFields
scanned note
Patient: J. Doe
BP 120/80 — stable
Follow up in 2 weeks
— Dr. M. Reyes
{
"ocr": true,
"fields": 9
}
11

Handwriting & Messy Scans

Read handwriting, faded carbon copies, and photographed receipts with a vision model — where classic OCR returns garbage.

Extract
OCRVision
lang: ar
مرحبا بكم في الشركة
التقرير السنوي ٢٠٢٦
税务报告 · 2026
{
"language": "hi",
"blocks": 18
}
12

Documents in Any Language

Detect the language automatically and OCR across a wide range of scripts — Arabic, Chinese, Cyrillic, Hindi, Japanese, and more.

Extract
OCRMultilingual
patient record
Name:
SSN:
Card:
Diagnosis: stable, routine follow-up.
{
"redacted": 14,
"categories": 4
}
13

Redact Sensitive Data

Detect personal information across many categories and mask, hash, or tokenize it — with a report of what was found and where.

Enrich
CompliancePIIRedaction
42-page report
{
"summary": "…",
"key_sentences": 6
}
14

Summarize Long Documents

Turn 80-page contracts, reports, and filings into a tight, grounded summary — extractive or abstractive.

Enrich
KnowledgeSummarization
lang: ar
مرحبا بكم في الشركة
التقرير السنوي ٢٠٢٦
税务报告 · 2026
{
"source": "de",
"target": "en"
}
15

Translate Documents

Translate extracted content into another language while keeping structure — headings stay headings, tables stay tables.

Enrich
KnowledgeTranslation
incoming doc
type: Invoice → AP
{
"label": "Invoice",
"route": "AP queue"
}
16

Classify & Route Documents

Answer “what is this?” automatically — invoice, contract, complaint, resume — and route each document to the right queue.

Enrich
IDPClassificationRouting
contract clause
Acme Corp agrees to pay $50,000 to Beta LLC by March 2026.
{
"entities": 11,
"people": 4
}
17

Extract Entities (NER)

Pull the who, when, and how-much out of free-form text — people, organizations, locations, dates, and amounts.

Enrich
KnowledgeEntitiesNER
contract clause
Acme Corp agrees to pay $50,000 to Beta LLC by March 2026.
{
"keywords": 8,
"topics": 3
}
18

Pull Keywords & Topics

Auto-tag documents with their keywords and key phrases for findability — no manual curation.

Enrich
KnowledgeKeywords
physics paper
∫ f(x) dx = F(b) − F(a)
{
"equations": 5,
"figures": 2
}
19

Extract Equations & Formulas

Lift the math out of scientific and engineering documents as clean LaTeX, with its position on the page.

Extract
ResearchMathLayout
combined scan
Invoice
Contract
Cover letter
{
"documents": 3,
"boundaries": 2
}
20

Split Combined Scans

Detect document boundaries inside one big scanned PDF and split it into separate, labelled documents.

Extract
MigrationSplittingLayout
contract v2
The term shall be 12 24 months.
Either party may terminate with notice.
{
"revisions": 7,
"authors": 2
}
21

Extract Track-Changes & Revisions

Surface the tracked changes and revision history embedded in Office documents and PDFs — who changed what.

Extract
LegalRevisions
contract v2
The term shall be 12 24 months.
Either party may terminate with notice.
{
"added": 5,
"removed": 3
}
22

Compare Document Versions

Get a reliable diff between v1 and v2 — additions and removals highlighted, for prose and tables.

Extract
LegalComparison
POST /v1/jobs 202
{ "id": "j_8f2", "status": "queued" }
webhook → done
{
"job_id": "…",
"status": "queued"
}
23

Fire-and-Forget Processing

Submit a job and get a webhook when it is done — no holding a connection open per document.

Acquire
PlatformAsyncWebhooks
meeting.mp3
{
"segments": 47,
"speakers": 2
}
24

Transcribe Audio & Video

Turn meetings, calls, and webinars into searchable, structured transcripts — in the same pipeline as your documents.

Extract
MediaAudioTranscription
figure 3
Revenue by quarter ($M)
{
"captions": 4,
"figures": 4
}
25

Describe Images & Charts

Generate descriptions of figures, charts, and diagrams inside a document — so your RAG system can see them.

Enrich
MediaVisionCaptioning
shipping label
SKU-7741-AC
{
"codes": 2,
"type": "QR"
}
26

Read QR Codes & Barcodes

Detect and decode QR codes and barcodes found in document images — shipping labels, tickets, inventory sheets.

Extract
LogisticsBarcodesVision
research page
Transformers improved results[1] across many tasks[2].
[1] Vaswani et al., 2017
[2] Devlin et al., 2019
{
"citations": 9,
"footnotes": 4
}
27

Parse Citations & Footnotes

Pull references and footnotes out as structured data, and tie excerpts back to the claims they support.

Enrich
ResearchCitations

Open-source primitives, composed into one backend. Curated cohort of design partners. Apply to work with us.

Cookies

We value your privacy

Xberg uses cookies to improve your experience, personalize content, and analyze traffic. You can manage your preferences at any time.