Batch-scanned mail and stapled intake packets arrive as one file that's really many documents. Segment finds each boundary and splits the file into typed, page-ranged pieces, each tagged high, medium, or low confidence.
This is a single inbound PDF — a claims packet that arrived as one continuous scan. Segment reads the pages, marks the seams, and returns four labeled documents with the exact page range of each.
Pick how deep the seams go. Page-level split is the fast, cheap pass; block-level goes inside the page and hands back the parsed markdown.
Detects where one document ends and the next begins, without a full parse. It returns just the segment boundaries — page ranges and titles — and no OCR text, so it's the fastest and cheapest way to break a packet into its parts. Priced at $0.50 per 1,000 pages.
Lets a seam fall mid-page at a specific content block, guided by an optional custom prompt — "start a new segment at each invoice." It runs a full Convert alongside the split and returns the parsed markdown for each segment. Priced as convert + segment, $4.50 per 1,000 pages.
Hand Segment a batch-scanned PDF and it works out where one document ends and the next begins. No template to maintain, no separator pages to insert, no manual splitting step.
Every segment returns with a type label — invoice, contract, receipt — the pages it spans, and a confidence level. That is enough to route each one to the right downstream processor without a separate classification pass.
The packets you actually receive are scanned crooked, photographed on a phone, and half handwritten. Convert reads them first, across 90+ languages, so even a messy inbound PDF still splits cleanly.
Run Segment first in a Pipeline to split the packet, then send each component to Convert and Extract on the schema that fits its type.
A short schema names the section types worth looking for. Segment returns each match as a typed record you can loop over and hand straight to the processor that handles that type.
import json
from datalab_sdk import DatalabClient, SegmentOptions
client = DatalabClient()
# Name the segment types you expect. The schema tells Segment what
# to look for in a multi-document packet.
schema = json.dumps({
"segments": [
{"name": "invoice", "description": "Vendor invoice with line items"},
{"name": "contract", "description": "Signed agreement or T&Cs"},
{"name": "receipt", "description": "Proof of payment"},
],
})
options = SegmentOptions(segmentation_schema=schema, mode="balanced")
result = client.segment("intake_packet.pdf", options=options)
for segment in result.segmentation_results:
print(f"{segment['name']}: pages {segment['page_range']}")
# Example response:
# invoice: pages [0, 1, 2]
# contract: pages [3, 4]
# receipt: pages [5]Barcode and blank-page separators assume you control the scanner. The packets you receive from outside have none, so Segment reads the content itself to find where one document ends and the next begins.
Every segment comes back with the exact pages it spans and a confidence level, so you can route the confident splits automatically and review only the ones it was unsure about.
Splitting is rarely the goal on its own. Run Segment first in a pipeline, and the type it detects at each seam decides the schema that component runs against in Convert and Extract.
A single FNOL submission can carry a claim form, police report, repair estimate, and photos. Segment finds the boundaries; Extract handles each component.
See the deep-dive →MAILROOMBatch-scan the morning mail; let Segment split each PDF into its constituent documents before they hit your operations queue.
See the deep-dive →ONBOARDINGApplications that include W-9s, IDs, signature pages, and supplementary forms stapled together — split once, route by type, extract from each.
See the deep-dive →Start with an API key — nothing to host or operate.
Sign up →Run in-region, with no egress to US infrastructure.
Sign up →Runs inside your own AWS, GCP, or Azure account.
Talk to sales →Fully offline, the same model weights, dedicated support.
Talk to sales →Free tier, no credit card required.