Travel Tech Startup Converts Stalled AI Prototype into Production Booking Concierge with Claude on AWS Bedrock
At a glance
A Canada-based corporate travel technology startup had an AI travel concierge stuck at the prototype stage, built by a vendor no longer engaged. The prototype could chat about travel but could not book a trip, enforce a corporate travel policy, or run reliably in production, leaving the startup unable to show investors or prospective clients a working product in a crowded corporate travel software market.
NMD rebuilt the concierge as a production system on Claude on Amazon Bedrock, deploying it on Amazon ECS Fargate with automated flight, hotel, and rental car booking, real-time corporate policy enforcement, and receipt-based expense capture. The five-week engagement, funded through AWS Activate at no net cost to the client, replaced the stalled prototype with a live, testable product.
The startup can now demonstrate a working AI product to investors mid-fundraise and is already scoping the next phase: exposing the concierge inside other LLM’s to reach travelers without requiring a platform login, directly expanding its addressable distribution.
Industry
Use Case
Agentic AI, Workflow Automation, Compliance & Governance
Solution implemented
- Deployed Claude on Amazon Bedrock via AgentCore to power natural language travel request understanding.
- Hosted production services on Amazon ECS Fargate behind Elastic Load Balancing for reliability and scale.
- Used Amazon RDS for PostgreSQL to persist bookings, policies, preferences, and conversation state.
- Built booking, cancellation, and seat-selection tools calling the client's existing API layer exclusively.
- Added a policy-compliance tool that checks every booking against corporate travel rules in real time.
- Delivered receipt upload and OCR-based expense capture with CSV export for finance teams.
The value equation
- Converts a stalled AI prototype into a live, production travel concierge in five weeks.
- Automates flight, hotel, and rental car booking end to end without manual intervention.
- Enforces corporate travel policy automatically, reducing out-of-policy spend risk for client companies.
- Positions the platform as a credible AWS-native alternative to established travel management tools.
- Unlocks a new distribution channel via LLM app integration, reducing dependence on direct logins.
Company Snapshot
A Canada-based startup providing corporate travel and extended-stay booking, now building an AI-powered travel concierge to compete in the broader corporate travel management category.
Location
Canada
Customer Situation
The startup had already piloted an AI travel concierge, but the prior development team was no longer engaged, and the prototype could not move past basic conversation. It could suggest flights and hotels but could not complete a booking, apply a client’s corporate travel policy, or run at production reliability, security, and scale.
For a founder actively raising a funding round in a market crowded with established corporate travel platforms, a stalled prototype was a real liability. Investors and prospective corporate clients expect a working, policy-compliant booking experience, not a conversational demo.
This mirrors a common gap across early-stage travel and booking startups: strong product vision, thin production engineering capacity to turn an AI proof of concept into something clients can actually use.
NMD Solution
NMD reviewed the existing proof of concept and found it lacked the orchestration, policy enforcement, and infrastructure needed to move from demo to production. The solution required rebuilding the concierge on Claude on Amazon Bedrock with AgentCore-based tool orchestration, deploying it on Amazon ECS Fargate with Amazon RDS for PostgreSQL for persistence, and building dedicated tools for booking, cancellation, and real-time policy compliance checks against the client’s existing travel API.
What We Delivered
The concierge now interprets natural language travel requests and executes complete flight, hotel, and rental car bookings through the client’s own API, with no direct third-party integration required. Every booking is checked against configurable corporate travel policies in real time, and travelers can upload receipts for automatic expense extraction and CSV export. Within a five-week window, the team also delivered an approval workflow and long-term conversational memory, both originally scoped as stretch items, while a dedicated front-end UI remains a candidate for a follow-on engagement.
Ready to Transform Your AI Software Business?
See how New Math Data can transform your software business with AWS-powered innovation.