Imagine this: you're dreaming of your perfect vacation, and all you have to do is ask, "Where can I see whales in July?" No more endless internet scrolling. What if I told you there's an AI buddy who knows the top spots to witness these creatures and can even deliver ready-to-book travel suggestions just for you? Curious? Let me introduce you to a search bot and share how we've harnessed AI to make it happen.
so, what did we do?
Our mission was to make the LLM understand user intent, such as in the example "Where can I see whales in July?" In this case, we aimed to help the LLM grasp that the user intended to travel in July, locate where whales are, and recognize there's only one traveler. Before that, we created the list of presumed user intentions and embedded them, so that when receiving such a message, the LLM would look for these intentions in the message, understand it using properly designed prompts, and generate travel options. Then, we told the Large Language Model (LLM) the exact format we wanted the answer in. Overall, the user has more freedom in expressing themselves and communicating with the bot compared to traditional searches.
One challenge we faced was that the LLM could have missed important details when we wanted to feed all requests in one message. Therefore, we started breaking down the requests into smaller ones, which worked well. Additionally, having analyzed the user intentions with the end result in mind - the list of suggested trips - we programmed the search bot to ask a specific sequence of questions related to destination, trip duration, budgets, and more.
the outcomes, or benefits
- Improved User Experience
This chat bot doesn't just accept any request. It engages in a real conversation with users. And - unlike traditional search engines, this bot gets what you're saying thanks to properly engineered prompts that would make the LLM use the knowledge it has in a proper way. You can ask questions in plain language, like, "I want to watch whales in July." LLM knows how whales migrate and where they are in July, so it can suggest places to see them and even offer some travel options. If it needs more info, it'll ask – just like a human would do.
- Enhanced Engagement
This chat bot isn't just about generating results. Instead, it keeps the conversation flowing and the users in. You can ask it all sorts of questions, and it'll respond with answers that match your needs, increasing the customization and travel impressions as well.
- Say No More: Automation
Who likes manually searching through endless websites for information? Or calling travel agents? Not us and not lots of people, especially the millennials and Gen Z. With this chat bot, you can have a full-scale conversation, and it'll dig up the information you need in a sec. It's like having a personal assistant, but way more affordable, knowledgeable, and fast. Plus, it’s available 24/7.
Here is a short video-recording for you to get an idea of what we've done:
the challenges: dates, dates, dates
Dealing with dates was like navigating muddy waters. Users can be quite creative with their date requests – some are specific and some don't mention dates at all. Some use vague terms like “the next season” and you should figure out how to define the “today” and understand what the next season is like. Lots of versions, indeed. We had to find a way to turn all that into a solid instruction for LLM to understand all the cases and to be able to transform any user text about dates to a standard date object that can be used in search queries.
how the LLM comes to play
Well, LLMs nowadays can understand your requests on a whole new level. They can figure out what you really mean, even if you're not super clear. For instance, if you say, "I want to see penguins," it knows you're probably talking about Antarctica. It doesn't just sum-up keyword-based results. It thinks deep, builds associations, and gives you sophisticated answers.
Oh, and we told them not to make stuff up. So, if there's no good answer, it won't try to bluff.
the bottom line
AI is changing the game in the travel industry and this search bot is a good reference. It's taking user experiences to a whole new level, making searches engaging, automating tasks and leaving room for more complex tasks. Sure, we had to tackle challenges like handling dates and many more, but the benefits pay off. As the travel industry gets back on track, there would definitely be more AI-driven solutions like this, making travel planning smoother than ever.