ItinerAI

Planning trips un-necessarily hard. Drowning in browser tabs, creating complicated itineraries? How is any trip supposed to make it out of the group chat. But its 2024, now there's an app for that.

Comment

Inspiration

We often dream of creating the perfect itinerary for an upcoming trip, but the reality of planning can be overwhelming. The need to cross-reference various websites, compile findings, and share a cohesive itinerary with friends requires a significant time commitment. This process can be even more daunting when a minor change necessitates numerous updates, disrupting the entire schedule.

What it does

Our app provides a seamless experience by allowing you to effortlessly view your itinerary items on an interactive map, complete with directions. With the help of our intelligent chatbot, you can receive personalized location suggestions and easily add new destinations to your itinerary. Additionally, you have the flexibility to manually rearrange and edit locations, ensuring your travel plans are perfectly tailored to your preferences. This integration makes your journey both enjoyable and efficient, giving you full control over your itinerary with ease.

How we built it

We used REACT with typescript ad node JS as backend. The frontend is deployed using an Azure Static WebApp, which hosts this node JS server as an Azure Function App. The node server talks to our Container App, which hosts a LangChain agent connected to our Azure Cosmos Mongo DB and the AzureOpenAI agent. The node server is able to take all the information back, update the directions using the google maps API, and then serve the data to the front end.

For more details and a digram, see the end of the demo video!

Challenges we ran into

Handling data imports: Where we need to figure out what information are important and what are not for the model to produce the correct suggestions without introducing unnecessary info that might confuse the model.

Too many tokens: Where responses from the chatbot often times contain way more information than we actually need. Hence causing us to easily exceed the token limit.

Formatting the response: Trying to specify the exact format we want the chat bot to provide. Which often times does not work as any slight modifications will cause the json parsing to fail.

Understanding before, after, start, end: Getting the chatbot to understand the relative order of the data, and insert itinerary locations appropriately.

Accomplishments that we're proud of

Able to utilize LLM to interact with our database and have seamless integration with Google Maps. Getting the whole thing deployed and running on Azure was a challenge, but once it was working was super rewarding!

What we learned

CI/CD pipelines, creating resource groups on Azure, using RAG to connect database with chatbot. Azure OpenAI. We feel confident to integrate and create chatbots into any applications!

What's next for ItinerAI

Better memory system and context for the chatbot, more exact result, better user experience, i.e auto completing the prompts to add to the itinerary. More locations than Tokyo!

Built With

Share this project:

Updates