10+ Years In Business | 4 Continents |
16+ Countries | 32+ Locations

Contact Us

GenAI-Powered Destination Assistants

Project Overview

A travel-tech platform for destination marketing organizations (DMOs), partnered with Athenaworks to scale a Retrieval-Augmented Generation (RAG) platform that powers 80+ AI-driven virtual assistants across global destinations.

By combining custom vector databases, agentic workflows, multilingual handling, and a highly cost-effective LLM strategy, the solution empowers their clients; cities like San Francisco, Atlanta, and Puerto Rico; to deliver real-time, hyper-relevant guidance on restaurants, events, and experiences.

The results: 8x faster deployments, 7x platform growth, near-zero engineering lift for new cities, and unmatched customizability in a space where 80% of GenAI projects still fail to reach production.

Industry

Travel Tech

Tech Stack

Cloud computing: AWS (Kubernertes /EC2) 

Orchestration: Github actions

Vector Store: Custom PostgreSQL with PGvector extension

Pipeline Control: DVC

Language Model: GPT-4o Mini (fallback: GPT-4 for reasoning)

Front-End Config: Internal “Model Builder” tool for Sales/Product

Scraping Infra: Bypass tools for Cloudflare-protected sites

Evaluation: LLM scoring + internal QA processes

Client Background

A travel-tech platform that partners with destination marketing organizations (DMOs); such as cities, regions, or tourism boards; to enhance visitor engagement. Unlike online travel agencies that focus on bookings, it’s core value lies in delivering personalized and revenue-aligned destination guidance through AI. In 2023, the company underwent a significant rebranding, reflecting its new vision as an intelligent AI-powered destination concierge.

The Challenge

The client faced multiple technical and operational challenges:

  • Content Overload: DMO websites often contain thousands of URLs and resources, overwhelming users looking for timely or relevant info.
  • Generic Recommendations: Open models like ChatGPT may recommend unrelated places, for example suggesting Seattle instead of San Francisco.
  • Revenue Leakage: Without alignment to business goals, GenAI answers could steer users away from sponsored listings or partners.
  • Scalability and Customization: Serving 70+ clients meant adapting the same GenAI core to local contexts, content, tone, and user expectations
  • Latency and Deployment Bottlenecks: Early iterations took up to 15 days to deploy a new model, limiting agility.
  • Filter out outdated or irrelevant content, such as expired event promotions

The Solution:

The Solution

Athenaworks built a multi-tenant GenAI concierge solution using the following components:

1. Retrieval-Augmented Generation (RAG) with Custom Vector Stores

  • Used custom PostgreSQL-based vector databases
  • Data scraped from public content on DMO websites; clients can include/exclude specific pages, languages, or URLs.
  • Built using Kubernetes + AWS, with Dockerized pipelines and DVC (Data Version Control) for artifact management.

2. Agentic Architecture & Event Awareness

  • Introduced heuristic + LLM filtering to exclude expired events (like  festivals, marathons) during model construction and inference.
  • Implemented agent chains (Chain-of-Thought) for reasoning over time-sensitive data and routing requests dynamically.
  • Future-ready: API integration for real-time event data ingestion

3. Dynamic Language Support

  • AI assistants can respond in any input language, including Mandarin, Spanish, or Thai, without prior configuration.

4. Business-Aware Response Prioritization

  • Built-in logic for content override: e.g., if a city wants to prioritize 10 sponsored Michelin-starred restaurants, the model routes users accordingly; even if the raw query suggests something else.

5. Model Building Platform (Self-Serve Interface)

  • Created a no-code model configuration tool for the Sales/Product team.
  • Allows internal teams to spin up new city models using templates—no engineering required, saving weeks of development per client.

6. Invisible Compliance & Page Scraping

  • Solved a major challenge in web scraping: most tourism pages use anti-bot protections from CDNs like Cloudflare.
  • Built internal tools that bypass these blockers without client technical lift, reducing friction and speeding go-live time.
Download PDF
Contact Us!
Linkedin-in X-twitter Instagram

©2025 CANWILL ATHENAWORKS HOLDINGS LLC