Worth It Dating

Semantic search users & read reviews before dating

Comment

Inspiration

  • We believe selecting your life partner is one of the most, if not the most, important decision in your life. However, people constantly resent and face negative experiences on dating apps, today’s most popular method of finding a life partner. This is why exploring this area is crucial, given its significant impact not only on our lives but also on those of many people we know.
  • Half of Americans that use dating apps describe their experience as negative according to a 2023 Pew survey. This is attributed to (1) widespread negative behavior (ex: 41% of users have experienced “toxic behavior”, including ghostings and catfishing) and (2) challenges in finding the right person (ex: 88% of users reported to be disappointed in users they have seen on dating apps). We have also witnessed and heard countless horror stories about these negative experiences from friends – further emphasizing the need for an alternative dating platform. We have also validated the value of our platform through dozens of interviews as we created the MVP.

What it does

Semantic search users & read reviews before dating

We rely on reviews for everything – why not for your next date? Current dating apps are like the Wild West – it’s impossible to know the true intentions and behavior of potential life partners, forcing users to scour through countless jerks and incompatible matches in an unvetted dating community. Our solution is a dating platform for serious relationships where you can Google search for your ideal partner and learn more about them through community reviews and ratings. We believe these two features enable the community as a whole to have higher quality matches and reduce negative experiences:

  1. Google search bar: We use a multimodal approach to create vector embeddings of all user profiles. Users can “Google” (semantic search) for people with specific traits or preferences by inputting a natural language query. For instance, a user can search for "Someone who’s entrepreneurial and likes to travel." Our semantic search would return relevant profiles, such as users with startups in their job descriptions or those with multiple photos in different countries. This empowers users to search for things important to them rather than navigating through arbitrary suggestions, significantly improving compatibility of prospective matches.

  2. Reviews and ratings: When you look at a user’s profile, you can see community reviews and ratings about them. For heterosexual relationships, women can leave a review/rating but cannot be reviewed themselves. For other relationships, both individuals can leave a review for each other after they match. This allows a user to see a person’s “red flags” and other details before they match, which decreases the chances of an incompatible date or negative experience.

How we built it

We started ideating on Valentine’s Day this year and launched the MVP on the iOS App Store in two months. We conducted dozens of interviews, which all had positive feedback. We have a stack-ranked roadmap that should allow us to publicly launch the dating platform over this upcoming summer.

Tech

  • Our website is created using Next.js and hosted on Vercel.
  • Our iOS app is created using SwiftUI, Firebase (Auth, Firestore NoSQL database, Cloud Functions, Cloud Messaging, Blob Storage, etc), and Pinecone vector database. We use LLM models from Vertex AI and OpenAI, but have explored other multimodal offerings.
  • We leveraged Gemini to speed up our development journey.

Challenges we ran into

Technical

We had difficulty identifying the most accurate embedding model for semantic search on both images and text of a user’s profile, since a user’s profile photos and their text biography should be in the same vector space. After experimenting with multimodal offerings from Meta, OpenAI, AWS, and Google, we ultimately decided to use Vertex AI to get text descriptions of each user image. We faced issues with the token size of multimodal embedding models. We ended up creating multiple embeddings for each profile and then averaging them to create a single profile embedding. This approach improved the accuracy of the semantic search and the default profile suggestions. We noticed that negative sentiment was difficult to capture during the semantic search process. For example, a user who searches “Someone who loves tennis” may retrieve a profile who “hates tennis” as a closer match than someone who does not mention tennis. We mitigated this challenge through prompt engineering to properly capture the semantic meanings of user inputs.

Business

As with all dating platforms, the main bottleneck in growing a new network to critical mass is getting enough female users on the platform (assuming majority of relationships are heterosexual). Therefore, the initial focus is on acquiring as many female users as possible through our platform’s features and marketing. (1) Women can read and leave reviews but cannot be reviewed themselves, empowering them to filter potential matches for safety and compatibility. (2) We plan to partner with female-centric brands such as Sephora and L'Oréal to target more women to improve engagement (ex: $5 Sephora gift card for 1st match). We can leverage referral programs where both current users and new joiners benefit though premium access to the platform and gift cards of our partners. We also aim to provide integrations with SMS and Whatsapp to unlock a whole set of users that do not regularly use dating apps. In order to reach critical mass as quickly as possible, we expect high customer acquisition costs at the start. According to Liftoff’s 2022 Dating Apps report, the cost per registration of a new user on dating apps is $5.26. Given that we have our core features ready for production and assuming minimal external costs, the prize from Hackathon can be used to acquire users.

Accomplishments that we're proud of

In just a few months, we learned a lot about LLMs and gained lots of hands-on experience with Gemini. We were able to go from an idea to MVP in just 2 months and get the app out on the iOS App Store's TestFlight. We are building the first dating app for serious relationships where people can search in different modalities for traits and things important to them.

A month ago, we had our Alpha launch with dozens of people. We led focus groups to learn more about the unique needs of women, LGBTQ community, and people of different ethnicities. We got amazing feedback from everyone about our product and lots of feature requests to add to our roadmap and backlog. We are proud to build an inclusive product for everyone.

Picture from Alpha Launch

What we learned

Technical

  • We became proficient in the iterative process, from coming up with an idea to publishing an MVP to the App Store.
  • We learned how to gather user feedback and research through hands-on discussions with users.
  • We figured out how to create an iOS app in SwiftUI using XCode.
  • We gained more expertise in cloud functions and other Firebase offerings.
  • We became more efficient using Gemini / Bard chat.
  • We leveraged third party offerings, including Pinecone’s vector database.
  • We followed best practices to maintain a codebase that's used by more than one person.
  • We are always learning more about multimodal embeddings, given the constant advancements in LLM technology.

Business

Current dating platforms don’t give users the ability to control which profiles to view beyond basic preference filters. Although this mechanism keeps users on the platforms for longer, this leads to a negative user experience due to delays in matches and unsatisfactory dates. As a result, users are always looking for new and better platforms to meet their needs. Our Google search allows users to find exactly the people they are looking for based on things important to them.

Moreover, current dating platforms also make it difficult to assess a future life partner before matching with them. Since profiles are self-made, it’s impossible for users to know the true nature and personality of a prospective match. Although current dating platforms purposely do this to increase user engagement at the cost of user satisfaction, we believe our alternative approach allows for high user satisfaction without having to sacrifice user engagement. Our community reviews and ratings provide more transparency about a person before meeting them, reducing negative experiences and optimizing for compatibility.

What's next for Worth It Dating

Improvements: We want to improve our two core features:

  1. Google search bar: Support more modalities like audio and video. Improve the quality of search by optimizing the embedding process, semantic search algorithm, etc.
  2. Reviews and ratings: Leverage these reviews to remove bad actors and prioritize good users, consistently maintaining a kind community.

Next Features: There are several features in our backlog; we plan to do more user research and appropriately prioritize them in our roadmap:

  1. Encompass the entire relationship lifecycle: After a user finds their significant other, they remain on the platform to ensure their relationship stays healthy.
  2. In-person date: Recommend in-person date ideas based on user profiles; opportunity to partner with brands/businesses.
  3. Verification: Identity, university, employer, income, etc.
  4. Messaging: Integrate with WhatsApp and text so that people can get matches without ever having to download the app. Supplement with an AI dating coach to help people with conversations.
  5. Social media integration: Integrate with social media platforms to get more user data to improve quality of suggestions.

Built With

Share this project:

Updates