Simplifying Vacation Planning with AI Travel Assistant by Al

Background

The AltexSoft team created an AI-powered travel assistant that understands natural language queries and simplifies vacation planning. The idea was conceived and realized as a proof of concept during an internal GenAI hackathon, where the project won first prize.

Like a traditional online travel agency, the assistant enables users to search for hotels and filter options by date, price, star rating, and other parameters. However, all interactions happen via chat, mimicking a live trip advisor. If a customer is unsure about their destination, the assistant provides personalized recommendations based on the context of the conversation.

Challenges

Built on the GPT-4 API and AltexSoft’s accelerator—a booking engine framework for travel providers—the assistant currently supports hotel and activity searches across 15 cities. The system can be expanded to include a broader search base and booking functionality, addressing key travel industry challenges.

Speed up time to market for travel booking software

Automate repetitive tasks to reduce operational costs

Provide 24/7 customer support

Enhance user experience with personalized recommendations

Value Delivered

Although still at the proof of concept stage, the assistant has significant potential to improve the travel booking experience.

Seamless, human-like conversation

The assistant engages users naturally, asking about interests and offering tailored recommendations. For instance, if a traveler says, “I want to go shopping, but my husband prefers the beach,” the assistant suggests three cities that match both preferences and explains the rationale behind each choice.

Refined LLM outputs

Since GPT-4 is trained on general data, it sometimes struggles with domain-specific queries. To improve accuracy, AltexSoft integrated retrieval-augmented generation (RAG). The system first searches a vector database containing curated travel data on 15 cities, retrieves relevant documents, and feeds it to the LLM for further processing. A context layer preserves conversation history, ensuring continuity in interactions.

Search for the most relevant hotels and activities

Once a destination is chosen, the assistant collects missing details (e.g., check-in/check-out dates, number of travelers, ages of children) and performs a search via AltexSoft’s booking engine framework API. It retrieves up to 20 hotels, highlighting the top three based on the query’s context. Users can modify their search—for example, requesting only five-star hotels or budget-friendly options. After a hotel is selected, the assistant automatically generates a personalized activity plan for the stay.

Approach and Technical Info

The proof of concept was built in just a week and involved a UX/UI designer, solution architect, front-end developer, and back-end developer. AltexSoft’s booking engine framework (i.e. accelerator), which includes pre-built hotel API integration, dramatically sped up the development.

The tech stack includes OpenAI API, LangChain for building RAG, Elasticsearch as a vector database, React for the frontend, Python for the backend, and WebSockets for two-way communication.

PythonReactWebSocketElasticsearchOpenAI API