CONCEPT
«Together» — a dating application designed to connect people not only based on physical attraction but also on compatibility of lifestyle, interests, and plans. The app’s main goal is to help users move beyond endlessly scrolling through profiles toward real shared experiences and connection through a common activity.
DATA SOURCES
OkCupid Profiles
— A database of dating service user profiles OkCupid Profiles is a public dataset comprising approximately 60 thousand user profiles from OkCupid. It includes data on age, gender, orientation, education, occupation, location, habits, interests, and users' self-descriptive text responses. This dataset has been used in academic research and is suitable for studying behavior and profile structure in dating. For «Vmešte» (Together), this source is useful not for directly copying users, but for investigating preference structures and building a prototype for a recommendation system. Based on this data, one can analyze which profile parameters frequently appear together, how interests are best grouped, and what characteristics should generally be asked during registration. For example, analysis shows that users often list interests in music, film, and urban culture simultaneously. In «Vmešte,» this allows us to avoid forcing a person to manually select dozens of tags; instead, we can suggest a pre-defined interest group, «Culture and Events, ” within which we can immediately recommend concerts, movies, and exhibitions.Meetup API
— Community and Collaborative Activity Data The Meetup API strongly aligns with the concept of «Together» because Meetup already focuses on events, groups, and people united by shared interests. Through the API, you can access data regarding events, their descriptions, categories, and groups. Starting in 2025, Meetup transitioned its API to GraphQL; full access to creating new OAuth integrations is now tied to Meetup Pro and platform approval. For «Together, ” this source can be used as a benchmark for building a system of themed activities. Even if you do not integrate Meetup directly, data on popular meeting formats helps inform which categories to build within the application—such as walks, board games, language clubs, sports, concerts, professional meetups, and more. For example, if the number of board game Meetup events is growing in a specific city, „Together“ can automatically boost the corresponding category and display a block to users titled „Looking for a Board Game Group.“ This way, external data helps the application respond more quickly to the real interests of its audience.Ticketmaster Discovery API + Foursquare Places
For populating the application with real reasons to meet, the Ticketmaster Discovery API and Foursquare Places API are particularly useful. Ticketmaster allows searching for concerts, festivals, sporting events, artists, and venues by location, date, and category; its Discovery API covers hundreds of thousands of events across various countries. Foursquare, in turn, provides data on cafes, restaurants, parks, museums, and other locations. The Places API supports searching by location, categories, and keywords, and its database contains over 100 million Points of Interest (POIs). This allows the dating mechanic to be enhanced with specific, safe, and suitable meeting spots.Example: Two users «Together» indicated an interest in indie music. Ticketmaster shows a nearby concert on Friday, and Foursquare helps select a cafe near the venue. The application might then present a ready-made suggestion: «You both love indie. There’s a concert Friday at 7:00 PM — want to meet at a nearby cafe beforehand?» This transforms matching from a simple person recommendation into a complete meeting scenario.MACHINE AND HUMAN PROCESSING SCHEMA
The «Together» concept is built on a division of responsibility between the human and artificial intelligence. The user remains the primary decision-maker: they complete a profile, specify interests and relationship goals, define desired meeting formats, rate recommendations, and ultimately decide whom to connect with and meet. Thus, the human is responsible for the context, personal preferences, feedback, and final selection. AI performs the analytical and recommendation work. It synthesizes profile data, in-app behavior, interests, external events, and available locations, subsequently searching for compatible options and suggesting suitable people, events, and meeting scenarios. As the application is used, the recommendations are refined based on the user’s actions and feedback.
Original size 3405x1744
The final model works like this: a person provides a direction → the AI processes the data and suggests options → the person makes a selection → the system learns from the outcome.
GUIDELINES
Guidelines for Automated Services in the «Vmeste» System. Rules for AI, recommendation algorithms, data collection services, and automated content generation.
Principles for Working with Primary Sources
1. Utilize verifiable sources—official APIs, event websites, institution databases, and data provided directly by the user. 2\. Record the source and the date the data was obtained, especially for events, schedules, pricing, locations, and other rapidly changing information. 3\. Do not substitute missing data with assumptions—if information is unknown, the system must indicate uncertainty. 4\. Cross-reference information from multiple sources when the accuracy of the recommendation depends on it. 5\. Adhere to user privacy and consent—personal data must only be used for the stated functions of the application and within acceptable limits.
Principle of Scientific Analysis
1. Distinguish real patterns from random coincidences and avoid drawing conclusions from isolated user actions.
2. Consider multiple factors simultaneously: interests, activity level, dating goals, time, distance, and feedback.
3. Regularly test the effectiveness of algorithms using measurable metrics: responses, matches, transitions to dates, and recommendation ratings.
4. Do not equate correlation with causation—the popularity of a specific activity does not automatically mean it improves dating quality.
5. Monitor for systematic biases and skew to ensure the algorithm does not discriminate against users or create overly narrow «information bubbles.»
Audience Adaptation Principles
1\. Consider the user’s goal—relationships, friendship, attending an event, or a shared hobby all require different recommendations. 2\. Personalize suggestions gradually, basing them not only on the questionnaire but also on actual behavior and feedback. 3\. Account for context: city, time of day, day of the week, event availability, and preferred distance. 4\. Do not restrict the user to overly narrow recommendations—some results should help them discover new interests and people. 5\. Give the user control over personalization—the ability to modify interests, hide topics, and explain why a specific recommendation was shown.Material Formation Principles
1. Provide the user with a complete action scenario, not just a set of data: person + activity + location + suitable time.2. Place the most crucial information first—what is being offered, why it is relevant to the user, and what the next steps are.3. Avoid information overload in cards: details should be available upon request, keeping the main screen clean.4. Separate facts from recommendations—official event data must not be mixed with AI assessment without clear designation.5. Maintain a consistent presentation style: concise headings, clear categories, visual cues, and a specific call to action—such as «Find a Company, ” „Join, ” or „Suggest a Meeting.“Original size 3405x1744