Inspiration
First, as a college student, I often find myself alone in the dining hall, watching everyone else surrounded by friends or partners. I want to talk, but I don’t know who to approach. Even if I try, how many people must I talk to before finding someone I truly connect with? Most of the time, my social energy runs out before that happens. It’s frustrating, and makes you wonder if something is wrong with you. The desire to talk is often spontaneous—this is why people don’t always use apps like Tinder or Eventbrite. You want to talk now, but no one's the right fit. What if an algorithm could help?
Second, we noticed that conversations sometimes flow better with three people, not two—especially when one acts as a “conversation driver.” That role—observing, bridging gaps, and guiding topics—has always been a rare human trait. But now, AI can take that role. In a time when people are glued to their screens and less comfortable with face-to-face talk, VibeEase uses tech to help people relearn how to connect in real life.
Third, beyond interest-based matching, VibeEase hides visual or demographic cues like gender or race. One teammate, Aaron, grew up in Asia and initially only felt safe talking to fellow Asians at college in the U.S. It wasn’t until a mutual friend introduced him to people of other backgrounds that he broke out of that shell. That’s what VibeEase can do—act as that mutual friend, breaking down the invisible walls that separate us.
What it does
VibeEase connects people nearby for real-time, in-person conversation. Users can browse nearby “conversation circles” based on listed topics and join one that interests them. Matching is based on user input (topics you like or want to avoid) and enriched by personal interest data pulled from platforms like Spotify, Netflix, Amazon, and Pinterest through Knot API. Gemini AI processes this data to generate a personalized interest profile in the form of a few key identity-defining keywords.
Once matched, the app (with both users' consent) records the conversation and provides live, AI-powered suggestions. These suggestions adapt based on recent dialogue and profile context—whether it's avoiding an unwanted topic, pivoting to mutual interests like “dog walking,” or offering small nudges that help socially hesitant users carry the interaction forward.
How we built it
We built the frontend with React Native using Expo, which significantly accelerated our mobile development process—this was our first mobile project. The backend processes user data and Knot API transactions, while Gemini AI summarizes this behavioral data into compact, formalized profiles to avoid large, inefficient prompts. We store and manage live conversations using MongoDB, dynamically updating the dialogue in real time and feeding only recent exchanges to Gemini for lightweight, effective topic suggestion.
The result is a smooth pipeline from interest extraction → matching → live support, all optimized for real-time, consent-based interaction.
Challenges we ran into
Real-time voice processing: Handling and responding to voice input in near real-time was complex. We used MongoDB to dynamically store dialogue, selectively including only the latest context in prompts to Gemini, reducing latency and prompt bloat.
Data overload from Knot API: User behavioral data from Spotify, Netflix, Pinterest, etc. was rich but massive. We developed a preprocessing step using Gemini to condense this into formalized interest profiles—abstract enough to be generalizable, yet personalized enough to guide conversation.
New platform learning curve: This was our first time developing a mobile app, but using Expo made development surprisingly smooth and enjoyable.
Designing a full-stack social solution: We realized that solving this problem required not just matching, but also sustained interaction support. Creating a two-layer system (matchmaking + in-convo support) required us to think holistically about user behavior, attention span, and emotional flow.
Accomplishments that we're proud of
We’re proud of building a functional prototype that allows users to find conversation matches in real time, start talking in person, and receive live conversation support—all in one seamless loop. Testers said they felt more confident starting conversations and were surprised at how natural the suggestions felt.
What we learned
We learned that AI isn’t just a content generator—it can play the role of a facilitator, helping people navigate social uncertainty. We also learned how to manage large behavioral datasets efficiently, and how to create AI guidance that respects human timing, emotion, and spontaneity.
What's next for VibeEase
Our next priority is building a robust relevance-based matching algorithm. We plan to represent each user’s profile—including their preferences, behavioral data, and expressed interests—as a structured string document. These documents will be embedded into vector space using language models, allowing us to compute proximity between users efficiently. We aim to implement this using Pinecone for scalable, real-time vector search, enabling smart and relevant matchmaking.
Easy Start With
“Hi everyone, this is a quick demo of our Hackathon project: VibeEase, an AI-assisted real-time conversation enhancer that helps people connect better in live events like conferences or dating meetups.
Our system is powered by Knot’s multi-platform mock data and API. We start by collecting a user’s behavioral data from apps like Uber, Netflix, Spotify, Doordash, Walmart, and more. For example, if someone often orders vegetarian meals on UberEats and watches health documentaries on Netflix, our backend uses a Gemini-powered model to extract their Key Profile—things like interests, habits, and conversation avoidances.
💡 Here’s the cool part: when you enter a room or an event, VibeEase matches you with someone based on location and time proximity—within 5 minutes and in the same city. Then, using your key profiles, our system intelligently suggests icebreaker topics and warns if a topic might be sensitive based on the other person’s preferences.
The system instantly generates their key profile: interests include yoga, wellness, and ambient music. When paired with the other User, who shares similar themes, we get a match with a conversation starter like “Have you tried any mindfulness podcasts lately?”
If during the chat someone brings up a topic flagged as avoid—for example, politics—our assistant can gently recommend shifting to a mutual interest and even suggest phrasing by the recorded conversation, like “By the way, I saw this cool yoga app recently...”
This makes real-time interaction more natural, empathetic, and aligned with people’s vibes—hence the name VibeEase.
In terms of location matching, we need to fine-tune how geographic proximity influences connections. We're currently exploring a balanced integration between manually inputted location descriptions (like “corner booth” or “north lounge”) and coordinate-based geolocation, ensuring both accuracy and flexibility for indoor environments.
By combining semantic relevance with physical context, VibeEase can better deliver spontaneous, high-quality social interactions—exactly when and where people are open to them.
Log in or sign up for Devpost to join the conversation.