The problem Bon Voyage solves

Problem

Feeling cooped up after all those lockdowns and the daily grind? You're not alone. Many people are craving a break from their routine. Travel brochures and websites tempt us with amazing getaways, but the hefty price tags often burst our vacation bubble.

Our Solution

We streamlined the process of planning by using LLMs, enhancing user experience. We achieved this by taking minimal user input such as the date of arrival, food preferences, health considerations. We also used realtime parameters weather and points of interest which inturn helps us to get better response than traditional means. All user has to do is fill up the form provided and ET VOILÀ! you will be given a really in depth itinerary which would've taken hundreds of hours of research or a thousand bucks to someones pocket.

Challenges we ran into

Backend challenges:

• Hallucinations (Ooooo Spooooooky... ): As the project uses LLM, it was prone to hallucinations from the start. This was solved by first implementing zero shot learning and then was improved upon by modifying it to few shot learning.

• gRPC (Life Saver): we had to create individual servers for our agents like weather,longitude and latitude etc. It was quite difficult to handle calls to all the different servers and would've costed a significant wait time, So we thought gRPC's API gateway would be our go to gateaway to manage our independent server.

• Redis (DON'T TOUCH IT, IT's FIXED... ig):LLM's are inherently slow. but we used redis, it cached our responses once for when we may have to call it again. it dropped our frequent response times from 1 mins down to 4 ms when cached.

Frontend:

• Auth0: used it for the first time, it was a bit difficult navigating the documentation. Keeping the session alive was difficult which was solved by using providers. Thankfully we were well guided by our mentors and ended up executing it successfully.