1 of 14

Berlin Bites

Final presentation

2 of 14

Agenda

​

Berlin Bites web app components and the contribution each track

The Work

Check out Berlin Bites - the restaurant recommender that gives you exactly what you are looking for.

The Solution

From the busy minds and hungry bodies of Berlin to an NLP-based restaurant recommender.

The Problem & �The Product

About regular team meetings, lost team members and the prioritization of value over functionalities

The Journey

3 of 14

Who We Are

UX Design Team

WD Team

Andrés

Deep Learning

Rashmi

Leo

Felix

Olimpiya

Pratima

Kristina

Mentors

4 of 14

Problem

Customers are overwhelmed by the amount of restaurant choices, leading to decision fatigue and unsatisfactory dining experiences.

Solution

Berlin Bites in an AI-driven restaurant recommender that mines user reviews and delivers personalized dining suggestions.

5 of 14

  1. Prompt based search: �Users describe their preferences in short text

​

  • Review Analysis: �Recommendations based on reviews.

​

  • Growing Database: �Let’s users add their own reviews.

Product Definition

6 of 14

Demo

​

​

7 of 14

Major wins & challenges

​

​

​

  • Lost team members
    • Only one DL techie left
  • Decision processes
    • All tracks start at same time
  • Lack of knowledge
    • Got lost in complex solution attempts without results
  • Actively seeking help
    • From mentor & track leads
  • Time management
    • Deployed 1 week early
  • Communication
    • Regular monday meetings

8 of 14

Web Development

​

​

​

  • Classic MERN app (Mongo Express React Node)
  • Mongo Atlas DB features 4200 Restaurants
  • 25 React components
  • Only custom CSS
  • Site is fully deployed (No mobile version yet)

→ https://berlin-bites-frontend.onrender.com/

9 of 14

Deep Learning

​

​

​

  • Leveraged SentenceTransformers to create meaningful numerical representations of restaurant reviews & user input
  • Applied simple machine learning model to identify restaurants closest to user description
  • Deployed the model as Flask API on pythonanywhere.com
    • Installed required python libraries
    • Connected frontend to API through flask-cors package
  • Extracted keywords from the reviews to describe each restaurant in a nutshell

10 of 14

User Research

​

​

​

  • ​
  • Easy and fast way to Find Restaurant
  • Make it easy for users to decide on a restaurant by displaying reviews, prices, and location.

Gains

Pains

What They Do

Insights from user interviews

  • Simply write a short text to find your desired restaurant.

​

  • Limited Familiarity with City
  • Too many Option to choose from
  • Difficulty to find specific Cuisines
  • Lack of reasonable Recommendations

11 of 14

User Test

12 of 14

User Testing Results

​

  • Search results disappear when moving from details cards back to main page
  • Users need to do this due to missing location information on main page
  • Users prefer to continue scrolling instead of clicking “show more” (both restaurants & reviews)
  • Simple, user-friendly and clean design
  • Banner suggests it’s a food website
  • Detail Cards present the information clearly and enable easy access to reviews
  • Map makes locating the restaurant intuitive

Pro’s

Con’s

13 of 14

Next steps & Future Potential

  • Improve recommendation algorithm
  • Expand the amount of reviews in database
  • Add tags for restaurants to the database
  • Add user profiles with user location & save/like functionality
  • Add book a table feature
  • Add share a restaurant with other users
  • Improve security of the website & model API

​

14 of 14

Thank you!!!