Strictly confidential

VFS Travel
Intelligence Engine

A trusted answer for every visa corridor, kept current at a cost that works.
Prepared for

VFS Global

Prepared by

360 Labs

Summary deck

September 2026

01The problem

Anyone can generate an answer

The hard part is maintaining a trusted answer across 40,000 corridors at a cost that works. Four things have to hold at once, and they compound at scale.
The frame
39,800

Corridors: 200 origins × 199 destinations, roughly 400,000 pages

VFS publishes
1,826

Corridors, 4.6% of the universe. Where production starts

Proven deeply
31

Corridors crawled end to end. Proof of mechanism, not of scope

Axis 1

Capability

Sources are scattered and change with no notice, no changelog and no API. Discovery never finishes.

Axis 2

Cost

Naive crawling and extraction at 40,000 corridors costs more than the answers are worth.

Axis 3

Governance

Models hallucinate. A named person still decides what reaches a traveller.

Axis 4

Ownership

The intelligence, the registry and the review decisions have to stay VFS property.

02 / 06
02What is live

What is live today

One 360 Labs engineer, a 6-hour sprint. Every passport-to-destination pair now answers, and each answer states how deep its evidence goes. Every count is computed from the data on disk.
Pairs answering
39,402
Full corridor answer 3.5%
Government rule 13.8%
Reference dataset 82.7%
Full corridor answers
1,369

Across 67 destinations, every line quoted from its source

1,338 read through the VFS content API, across 63 destinations and 161 origins. 53 destinations are 100% grounded.

Government-sourced
6,818

Pairs answered from a government or VFS source

17.3% of all pairs, and half of VFS’s own: 6,328 of the 12,672 pairs inside VFS’s 86 destinations.

Grounding, crawled set
99.6%

510 of 512 claims carried verbatim evidence

The 2 that did not were dropped automatically. Compression before extraction: 8.97M tokens to 6,718, with 16 of 16 probe facts recovered.

03 / 06
03How it works

How it works

A pipeline builds a verified knowledge base; a monitoring loop keeps it current. Answers are read straight from it, so serving costs nothing per question and is traceable to source.
01

Discover

Find, validate, qualify, then a human ratifies the source

02

Fetch

Cheapest rung first: HTTP, in-country IP, browser, stealth only if flagged

03

Harmonise

Rules where the source is typed, a model only where it is not

04

Ground

A verifier outside the model drops any line it cannot find verbatim

05

Knowledge base

A graph of facts with source, authority and fetch time

06

Serve

Database read. Zero model calls when a traveller asks

Monitoring loopDaily re-check of every page and PDF → material change only → re-extract → human review
Why cost holds

Effort tracks change, not corridors

At 400,000 pages: 94% unchanged after a week, ~5.7% hash change a day, and only 0.12 to 0.23% material, which is 480 to 920 pages and 16 to 31 reviewer hours a day.

Ownership

VFS owns the intelligence layer

Corpus, knowledge graph, source registry, prompts and every review decision. The model provider and the cloud are replaceable suppliers.

04 / 06
04Proof

The live prototype

Not a recording. Pick an origin and destination and ask; every answer carries the verbatim quote it came from.
chancery.360labs.tech/sandbox Live Open in new tab

Live prototype

The sandbox could not be loaded here. It may be offline, or you may have no network connection.

Open chancery.360labs.tech/sandbox
05 / 06
05Commercials

Commercials and next steps

One fixed build fee, one optional monthly fee. Infrastructure is bought by VFS directly and never marked up by 360 Labs.
Option 1 · Build and handoverOption 2 · Build and maintain
360 Labs build fee$26,000$26,000
Team and timelineA 4-person pod for 2 months: AI-native engineering lead, PM and 2 full-stack AI engineers.The same team, scope and timeline.
After the buildStructured handover; your team runs it.$2,500 a month for 2 engineers: maintenance, source growth and improvements.
Infrastructure, paid by VFS~$90 a year for the VFS estate, rising to ~$32,300 at full global coverage. Today’s running range: US$550 to 1,450 a month. Optional GPU cluster: ~US$11,000 one-time.
Next steps · all four can start this month

Gold-set review

You pick the corridors; we show the answers, quotes and sources.

Consular review

Your team checks jurisdiction and district handling.

Scheduled monitoring

Turn the daily loop on for an agreed set and watch what it catches.

Scoping call

Confirm priorities and access, then start the pod.

06 / 06
Live prototype
chancery.360labs.tech/sandbox
New tab
All slides
Arrow keys to navigate  ·  F present  ·  G all slides