AIRPLUSIntelligence

Under the hood

One pipeline, every piece of the trip.

Plus is not a chatbot bolted onto a search box. It is a pipeline that parses intent, searches flights, hotels and tickets in parallel, assembles one itinerary — and keeps working after the booking.

REQ
NL request
plain language
NLP
Intent parser
LLM extraction
INV
Inventory search
flights · hotels · tickets
RNK
Ranking + prediction
price · trend · taste
ITN
Itinerary assembly
one aligned plan
MON
Monitoring loop
continuous
RBK
Auto-rebooking
every component

the same request that books the trip arms the watch that keeps improving it

parser connectors ranker assembler watcher rebooker

Watch me think

A sentence in, a structured trip out.

This is the first hundred milliseconds of every conversation: the request parsed into the object that drives search, ranking, assembly and the watch.

intent parser · live tracesimulated
heard

route ✓dates + flex ✓cabin ✓hotel constraints ✓tickets ✓watch policy ✓
understood

The conversational trip agent

A sentence is a perfectly good spec.

"Quiet 5-star near the Louvre" carries real constraints — location, category, atmosphere — and Plus parses all of them. The request becomes a structured trip: routes, dates, cabins, neighborhoods, entry times, and the soft preferences most booking forms throw away.

Then the searches run in parallel — flights, hotels, and tickets at once — and the results come back assembled: arrival aligned with check-in, museum entry placed on the right morning, one total instead of five carts.

planning one trip35 min in
with Plus✓ one message — itinerary assembled
“4 days in Paris in March — flight, quiet 5★, Louvre tickets”
by hand2/9 sites, still going
flight search nº1flight search nº2airline sitehotel searchhotel reviewsmap checkmuseum ticketsshow ticketsspreadsheet
startdrag to scrub the evening2 hours

The price-watch loop

Booking is the beginning, not the end.

Everything in a trip reprices — fares constantly, room rates daily, even event tickets as inventory shifts. The monitoring loop treats a booked itinerary as a living object: every component watched continuously until the trip starts.

Price-prediction models decide what the watch does next: hold when the trend is easing, arm a strike when a dip looks probable, and check the fare rules before any move. When a better price appears in the same or better class, Plus doesn't notify — it rebooks, then reports what changed.

one watched fare · forecast · sim watching
30-day historynowmodel forecast
trend: easing over the windowholding — better likely

Preference & personalization models

It learns how you travel.

Two travelers asking for "Paris in March" want different trips. Preference models learn the patterns — lead times, cabin tolerance, the kind of hotel that gets a yes — and shape both the search and the ranking before results ever appear.

The same models tune the watch: what counts as "better" for a design-hotel loyalist is not what counts for a family of five, so the rebooking rules adapt per traveler rather than per route.

learned traveler segments · demo

how I adapt → surface premium-cabin options early, while the calendar is still open