Book a DemoSign Up
> technical_architecture

Built like infrastructure. Not like software.

A look inside the Propellus processing layer — how a visa application flows from a traveller's first message to a scored, validated, document-complete submission. Configured, not coded.

platform AI-native status active development spec updated Jun 2026
propellus · e-visa_assistant
Propellus AI Engine processing
I’ve read your passport. Confirm these details:
Full nameMARIA S. CRUZHigh
Passport noP1234567High
Expiry14 Aug 2031High
NationalityPHLMedium
87%
Application Strength Score87% likely approval · 3 fixes suggested
live · extract → validate → score
> intelligence_layer

The Propellus AI Engine.

Every application is the runtime execution of a published workflow. The engine does four jobs.

// extract

Read documents

Reads passport bio-page, photo, tickets, and supporting documents into structured, typed fields with per-field confidence.

// validate

Check rules

Checks every field and image against destination requirements. Flags problems and auto-corrects within policy bounds.

// score

Rate strength

Computes the Application Strength Score — approval likelihood with a per-dimension breakdown.

// orchestrate

Run the flow

Runs the conversation turn by turn, branching on prior answers, routing to suppliers and payment.

> the_workflow_model

A new visa journey is assembled, not coded.

The conversation a traveller has is built in a visual workflow builder — five node types, fourteen input widgets, bound to live platform data. Adding a destination or changing a government rule no longer takes an engineering deploy.

Step
ask destination
Condition
if Schengen
Message
documents prompt
AI Action
extract + score
End
submit + track
Propellus E-Visa Assistant
Where are you travelling to?
France
condition: Schengen → insurance required
Great — let's get your documents ready.
passport.jpg uploaded
Extracted & validated. Strength score: 87%.
submitted → tracking live
Draft → Preview → Publish, with versioning and rollback. A published flow can't go live with a broken reference.
> application_lifecycle

From first message to filed. A guided conversation.

01

Greet & qualify

The assistant greets by agent brand, confirms residency and destination, and matches the right visa variant.

02

Capture

Collects traveller details, passport, photo, travel dates, and supporting documents through a chat — not a form.

03

Extract & confirm

AI reads each document and surfaces the fields for the traveller to confirm. Nothing is accepted silently.

04

Validate & correct

Fields and photo are checked against destination rules; the assistant explains what it fixed or what still needs attention.

05

Score

The Application Strength Score shows approval likelihood, a breakdown, and a ranked list of concrete fixes — before submission.

06

Submit & route

Payment is taken, the application is routed to the right fulfilment channel, and real-time status flows back to the traveller and the agency.

> application_strength_score

Every traveller sees their odds before they submit.

Before submission, the engine scores the application against historical destination outcomes and shows exactly how to improve it.

Application Strength
// TRV-A92X
Document Quality
0.92
Travel History
0.78
Financial Profile
0.84
Purpose Clarity
0.71
Passport Validity
0.96
Profile Complete
0.88
→ Add 3 months of bank statements → Clarify purpose of visit → Re-upload a clearer photo
LIKELY APPROVAL 0 out of 100 · benchmarked READY
The score is decision-support, not a guarantee — calibrated, explainable, and never used to promise a government decision.
> how_the_ai_behaves

AI prepares the work. The human approves it.

Every AI surface across the platform follows the same principles.

Confidence is visible

Every output carries a High / Medium / Low indicator.

Reasoning is visible

Every suggestion includes a one-line "why."

Override is always available

Accept, edit, or dismiss, every time.

Audit is automatic

Every suggestion and decision is logged.

Privacy is preserved

One agency's data never trains another's.

The rule behind all five

No consequential action happens without explicit human confirmation.

> under_the_hood

AI-native architecture, from the ground up.

AI-native processing

Python monorepo with autonomous AI agents handling extraction, validation, scoring, and orchestration.

In active development · targeted Q3 2026

Cloud infrastructure

AWS and Azure with data residency options, built for government-grade compliance.

Security docs available on request

Document understanding

OCR-plus-structure models return typed fields with confidence, keyed to Document Hub schemas.

Targets ≥95% accuracy on passport bio-page
> what_we_re_building_toward

The Propellus Travel Score.

Every application generates a structured traveler record — identity, history, document quality, compliance signals. We're building the Travel Score: a real-time traveler identity graph sold to airlines, insurers, and governments as a data product.

Status: in design · targeted 2027
TRAVEL
SCORE
identity
verified
travel history
14 trips
document quality
0.94
compliance
clean
> developer_access

API integration for OTAs in development.

Targeted availability: Q3 2026.