LocalLens

Landing
Plan your trip
Dashboard

Problem

traveling is super exciting and fun, but also bewildering.

Where should I stay?

What neighborhood to visit?

What restaurants to try out?

What landmarks and activities are must-do's for me?

the internet and ai only helps you so far.

Whenever you book that AirBnb, you're worried about expectations vs reality.

Whenever you eat out at a new restaurant, you think about it living up to its reviews.

How are you supposed to know if any places are unsafe at night?

What we built

An app that helps connecting tourists to trusted locals through AI for crafting an authentic travel experience

Why we built it

helping make travel more magical, less perplexing.

Immerse yourself in the culture through the perspectives of a local.

prioritizing traveler safety.

According to a global report, 87% tourists were apprehensive about travel safety.

contributing to the gig economy.

Creating more opportunities for the 150+ million workers in the economy.

How does it work

Get on board

Quickly sign up with Google to plan right away.

Plan your trip

Choose the location and trip dates and provide a brief summary of what you want to do during your trip.

Get matched to expert locals

Based on your trip's location and your interests, you'll be matched to expert locals whom you consult over text, Zoom, or even in-person.

Enjoy!

Discuss your interests using your preferred mode of planning and enjoy your best trip yet!

How we built it

We built the web-based app on the node.js and express tech stack. For parsing interests from the bio entered by locals and the description for the trip, we're using Cohere AI's classification endpoint which takes a set of prompts and labels and trains itself to then predict what the input could have as labels. We also used Auth0's web authentication for Google sign-in and it helped us to make the authentication for our app quicker and easier.

Challenges we ran into

The Cohere AI's classification endpoint is made to classify only one label at a time, so it was difficult to extract more than one label from the given prompt. We fixed that by having an arbitrary confidence value for a certain label, so if it was above that value, we considered that label too. This ended up working perfectly as we tried out with different prompts.