Building AI Hotel Search for a Travel Booking Platform
DBB Software built an AI search that turns a traveler's written request into a structured hotel query, along with the customer-facing platform around it. It runs in production on a live booking site.
Industry
Technology
Service
AI Development
Team
5 Full-Stack Engineers
Project State
February 2026 - Ongoing
Country
NDA
About the Client
The client runs an AI content and search platform for the travel industry. Travel agencies connect their own hotel inventory to it, and the platform enriches what they hold, filling out thin descriptions, finding better photography, and confirming hotel facts.
The Client's Initial Request
The client engaged DBB Software to extend the AI search and build out the platform's customer-facing side after the search became the product's main feature.
Content Parity Across Agencies
Hotel descriptions and photography pass between agencies unchanged, and both go stale over time.
01
Search That Takes a Sentence
Travelers describe a trip in their own words, with dates, budget, party size and the kind of view they want. The search took filters.
02
Customer-Facing Platform Functions
Agencies needed billing, usage and statistics in a dashboard they could run themselves.
03
Inventory Inside AI Assistants
The platform's hotels needed to be reachable from ChatGPT and other assistant clients.
04
The Deliverables
DBB Software worked as a full-stack team across the AI pipeline and the platform around it. The work went into turning a written request into a query the existing inventory could answer, and into the customer-facing functions agencies run on.
Natural Language to Structured Query
A traveler types the whole request at once: hotels in Miami for next month with a beach view, two adults and two children, 8,000 budget.
The pipeline reads the intent, breaks the request into filters and tags, and matches hotels against them.
Results carry reasoning, so the traveler sees why a particular hotel suits what they asked for.
Hotel-name search runs on the same path, so a traveler who types a specific hotel reaches it directly.
Retrieval Over Agency Inventory
Agencies ingest their own inventory, and the platform processes and transforms it into the form the search reads. Content enrichment runs on the same data.
Related content is retrieved through embeddings held in vector storage, so a query reaches material that matches its meaning.
The pipeline runs mainly on open-weight models across several inference providers, which keeps the cost of a query under control.
Versioned Agent Orchestration
DBB built the orchestration layer from scratch. It runs multiple versions of a workflow and of individual agents at the same time, so a change to the pipeline can be introduced and compared against the version already serving traffic.
Evaluation Sets and a Scored Baseline
The team established a performance baseline and built evaluation sets from data resembling real user prompts. Pipeline changes are scored against those sets before they ship.
New agents, revised prompting and worked examples are each measured against the same baseline.
Search quality moves on a score, so an improvement is demonstrated before a traveler sees it.
Agency Dashboards and Connector Access
The customer-facing side gives each agency billing, usage and statistics for its own account.
The platform exposes its hotel data through MCP, the open standard assistants use to call external systems.
ChatGPT and other assistant clients query the inventory directly through that connector.
Results Achieved
Trips Described in Plain Language
A traveler types dates, budget, party size and the view they want, and gets matching hotels.
Reasoning Attached to Every Match
The search says why a hotel suits the request, so travelers compare on substance.
Measured Search Accuracy
Wrong results at launch were corrected by scoring pipeline changes against evaluation sets built from real prompts.
Differentiated Hotel Content
Fuller descriptions, refreshed photography and confirmed facts give an agency inventory its competitors have not already published.
A Storefront Adopted Whole
One agency replaced its own website with the platform. The AI search runs there in production.
Hotels Queryable From ChatGPT
MCP access puts the platform's inventory inside the assistants travelers are starting to plan trips in.
Have a Product Idea but No Technical Plan Yet?
Our Product Discovery turns a vision into a clear plan to build it through scoped requirements, the right technology choices, a sound architecture, and a roadmap that surfaces and de-risks the make-or-break unknowns before they cost you.
Contact Us
"Most of our work starts with a 30-minute call where someone describes a product they're trying to ship and one part of the engineering picture they can't get around.
If that's where you are, let's set one up; I'll tell you straight whether we're the right fit.”
Mina Morkos
Business Development Manager
Want a similar outcome for your team?
Ask our AI assistant — it can pull related case studies, talk through the approach, and put you in touch with the team if you want a deeper conversation.
