ClaudeAtlas

We make travel planning a breeze.

Comment 1

Inspiration

Travel is meant to be fun, but the itinerary planning process - going through endless travel blog posts and making tons of open-ended decisions - can be tedious and painful. We want to change that and make travel planning a breeze.

https://www.loom.com/share/23910f125e5248b196cf05335e1c4a34?sid=41df28ba-65ad-4cfe-b5f6-1110355476bc

What it does

  1. We make it fun to decide on travel ideas - choose between fun options that we come up with, and visualize the day's plan on both a calendar and a map.
  2. We allow you to easily iterate on the initial travel plan - use natural language prompting to replace a specific event or whole chunks of the itinerary with new ideas!
  3. We magically ship the final plan into your Google Calendar - just pack your bags and go see the world!

How we built it

  1. Frontend
    • Built with NextJS, delivering a seamless UX with ease of replacing itineraries we didn't like, right down to specific time blocks of the day
  2. Backend
    • Built with FastAPI
    • Utilised Langchain to chain multiple prompts to provide non-overlapping itineraries, and allowing users to provide location constraints
    • Processed translated unstructured Claude's brainstormed itinerary outputs into structured queries against Google Maps and Calendar APIs

Challenges we ran into

  1. Prompt engineering is tricky - we faced a challenge where the outputs of the LLM were fine on the Claude UI but were truncated when posted as an API request.
  2. Speed of generating ideas from the LLM is crucial for the user experience - we had to experiment with reducing the prompt input without compromising on the quality and consistency of the output.
  3. We also experimented with the placement and content of our directives to improve model performance, along with generating high quality examples

Accomplishments that we're proud of

Creating an easy-to-use, seamless end-to-end experience from brainstorming travel ideas to having the plan in your calendar app.

What we learned

Building an LLM pipeline to power the product, and delivering a superior user experience around the generated outputs.

What's next for ClaudeAtlas

  1. Optimising the prompts for improved performance (Accuracy and latency)
  2. Iterate on the UX by gathering user feedback
  3. Making itinerary editing experience more seamless by making Clause satisfy more sophisticated constraints
  4. Making it easy to share and edit itinerary with friends
  5. Integrating with 3rd party tools to complete the travel planning experience (Eg. booking of accommodations, flights)
  6. Bringing in latest events happening live in the city by integrating Claude with web-search capabilities
  7. Support multi-day trip planning

Built With

Share this project:

Updates