Document AI

Photos and PDFs in. Structured, checked data out.

A vision-language model reads travel documents; deterministic code validates what it reads. Built into the chat and available as standalone APIs.

Book a demo See pricing

Screenshot rate check Auto-route Passport Visa document Business card

Screenshot rate checkLive

Beat a competitor’s quote from a screenshot.

Send a screenshot of any flight or hotel offer. TravoPilot extracts it in a single pass, works out whether it’s a flight or a hotel and runs a live search in the same reply.

The problem: Customers send competitor screenshots on WhatsApp; agents re-type them into a search tool to compare.
How it works: One extraction pass detects the offer type and the details, including Arabic text, then triggers a bounded live search so the comparison stays fast.
Where AI helps: Vision reading of messy screenshots; the comparison itself uses live inventory.
Business value: Win price-sensitive customers in the same conversation they started.

Travel agenciesOTAs

Screenshot rate checklive

COMPETITOR SCREENSHOT

Detected hotel offerExtracted Dubai · 14–18 Dec · 2 adultsScript Arabic OCR ok

✓ Verified · live searchSame hotel and dates found live — your price is ৳ 2,140 lower than the screenshot.

Illustrative comparison

Image auto-routeLive

One attachment. The right reader.

A single upload endpoint classifies each attachment as passport, flight, hotel or other and routes it automatically. PDFs work on every surface through one shared rasteriser.

The problem: Users don’t say what they’re uploading — they just send it.
How it works: Classification first, then the specialised reader. Uploads are content-sniffed so a renamed file can’t masquerade as an image.
Where AI helps: Visual classification of the document type.
Business value: No menus or instructions needed — on web, app or WhatsApp.

Image auto-routelive

📎 attachment.pdfone file, no instructions

passport0.94

flight0.06

hotel0.03

other0.04

PDF rasterisedContent sniffing type verifiedRouted to passport reader

Illustrative demo · sample data

Passport readerLive

Read the MRZ. Verify the maths. Fill the form.

Machine-readable-zone first, with check-digit validation, then a visual-zone fallback and cross-check. The source and confidence of each extraction are always stated.

The problem: Typing passenger names and passport numbers by hand causes rejected bookings and costly name changes.
How it works: MRZ parsing with check digits; a VL-model fallback for the visual zone; a holder photo crop using face detection with rotation handling; personal-data and emergency-contact page fields; booking-aware when used mid-hold.
Where AI helps: The vision model reads damaged or angled documents; check digits and cross-checks catch its mistakes.
Business value: Correct passenger data in seconds. Private storage, short-lived links, masked numbers at rest and automatic deletion after 3 days.

Passport readerlive

PASSPORT · SPECIMEN P<BGDRAHMAN<<NUSRAT<JAHAN<<<<<<<<<<<<
EB01234561BGD9205148F3101207<<<<<<<<<<04

SurnameRAHMANMRZ ✓

GivenNUSRAT JAHANMRZ ✓

PassportEB0123456check digit ✓

Expiry07 Dec 2031check digit ✓

Specimen data · not a real person

Visa document scanLive

Saudi visa documents, structured.

Reads Saudi visa documents from PDF or image into structured fields, with private CDN copies and an admin console.

The problem: Umrah and Hajj operators process large volumes of visa documents by hand.
How it works: A dedicated reader with scoped API keys, so the capability can be given to a partner system without exposing anything else.
Where AI helps: Vision extraction of visa layouts in multiple scripts.
Business value: Faster Umrah group processing with fewer keying errors.

Visa document scanlive

VISA DOCUMENT · SPECIMEN

Visa typeUmrah✓

Holdermatched to passport✓

Validitystructured dates✓

Specimen · fields simplified

Business card scan & email draftsLive

From a stack of cards to partnership emails.

Reads name, title, company, emails, phones, website, addresses, social links, memberships and registrations from one to six photos, pairing fronts and backs automatically.

The problem: Trade-fair leads sit in a drawer because transcribing cards takes too long.
How it works: Back sides without a name are attached by email match; memberships written in native scripts are mapped (for example JATA); the company logo is cropped and verified as complete, with a private one-hour link.
Where AI helps: Fact-driven partnership email drafts per card, with a freshness gate that rejects invented numbers or claims.
Business value: Turn events into follow-ups the same day.

Business card scanlive

FRONT · BACK PAIRED

NameFarhan Kabir✓

CompanySkylark Holidays✓

MembershipATAB · IATAmapped

Logocropped · private linkverified

DraftPartnership email drafted from the card’s facts — no invented numbers or claims.

Fictional card · illustrative

Also as APIs

Use the readers in your own systems.

Passport, visa document and business card scanning are available as standalone APIs with their own scoped keys, per-key IP whitelisting for passports, and integration-reference PDFs.

Book a demo

See TravoPilot run on your inventory.

A tailored walkthrough with your markets, your languages and your booking flows. No slides — the live product.