Sorted

Inspiration

Every friend group has a chat. And every friend group knows the feeling. Someone drops a message, "what are we doing this weekend?", and suddenly forty replies later, nothing has actually been decided. Plans dissolve. Momentum dies. People give up and stay home.

This isn't a niche problem. It's one of the most universal friction points in modern social life, and it's particularly acute for Gen Z, a generation that is more connected than any before it, yet paradoxically struggles to convert that connectivity into real-world action. The irony is striking: the tools built to bring people together are often the exact reason plans fall apart.

Group chats generate enormous amounts of activity, messages, suggestions, reactions, counter-suggestions, but produce very little in the way of actual outcomes. Whether it's choosing a restaurant, planning a trip, settling on a movie, or figuring out a time to meet, the underlying dysfunction is always the same. The conversation becomes a substitute for the decision rather than the path toward it.

What makes this especially difficult for Gen Z users is the specific social dynamics at play. No one wants to be the person who imposes a choice on the group. Everyone is aware of others' preferences, budgets, and restrictions. The desire to include everyone, combined with the absence of any structured way to actually do that, creates a loop of endless discussion that quietly exhausts people and erodes the motivation to make plans at all.

Sorted is an AI-powered decision companion inside Zymix that helps groups move from conversation to action. Rather than asking users to adopt a new tool or leave the conversation they're already in, Sorted lives where the problem actually happens, working quietly within existing Zymix group chats, learning what each person wants, recognising when a group is stuck, and stepping in at exactly the right moment with a recommendation that genuinely works for everyone. Not just a guess, but a reasoned suggestion that accounts for real preferences, real constraints, and real context.

The inspiration for Sorted was simple: what if your group chat had a friend who always remembered what everyone liked, never got tired of the back-and-forth, and could always find the option that made everyone happy? That's what we set out to build.

Who it's for

Sorted is built for Gen Z social groups, specifically the 18 to 26 year olds who make up Zymix's core user base in the UK. These are people who are digitally native, socially active, and deeply reliant on group chats as the primary infrastructure of their social lives. They organise everything through messaging: nights out, travel plans, shared meals, spontaneous meetups, and they do it in groups, rarely alone.

What it is

Sorted is an AI-powered decision companion embedded inside Zymix that helps friend groups move from endless conversation to a clear, agreed plan, without anyone having to take charge, compromise unfairly, or leave the chat to figure it out elsewhere.

At its core, Sorted does three things:

It learns. Sorted builds a personal preference profile for each Zymix user over time, understanding dietary requirements, budget habits, favourite types of venues, locations they prefer, and decisions they've made before. This memory layer means Sorted doesn't need to ask the same questions every time. It already knows what each person is likely to want before the conversation even starts.

It listens. Sorted monitors the natural flow of group chat activity, not to surveil but to understand. It reads conversational signals: the back-and-forth, the unresolved suggestions, the growing message count with no outcome, and identifies the moment when a group has genuinely stalled. Rather than interrupting early or forcing a resolution, Sorted waits for the right moment to step in, making its presence feel helpful rather than intrusive.

It decides, together. When Sorted detects that a group is stuck, it surfaces a recommendation directly inside the Zymix chat. The suggestion is built by balancing every member's stored preferences and the specific constraints emerging from the current conversation. Crucially, Sorted doesn't just give an answer. It explains its reasoning, showing exactly why this option works for the whole group. It then invites the group to vote, turning a chaotic thread into a resolved plan in seconds.

The result is a group chat experience that feels the same as always, casual, social, low-pressure, but with one quiet difference: decisions actually get made.

How we built it

We built Sorted in three stages, and the most important one came before we wrote any code.

1. Thinking it through with AI, before building. We didn't start from a feature list. We started by reasoning through the problem out loud with Claude: which track to commit to, what scenario to focus on, and crucially, what the underlying logic of group coordination actually is. We broke the whole flow down into its real steps, from someone suggesting dinner to the group actually showing up, and worked out where the real friction lives. That's how we landed on our specific scope: not group decisions in general, but the "where do we eat tonight" deadlock, and not just picking a place, but breaking the social standstill. Narrowing down this carefully meant every later decision had a clear reason behind it.

2. Turning the agreed logic into a spec. Once the scenario, the features and the full user flow were locked down, we had Claude Opus 4.8 turn those agreed details into a structured Markdown specification. Because we had already thought the foundation through, the spec was clear and concrete rather than vague, which made everything downstream more organised.

3. Building and iterating with Claude Code. Our main development tool was Claude Code. We dropped the spec into it to scaffold the project locally and generate a first working prototype. From there we iterated tightly: refining the AI intervention message, adding a transparent voting card so the group's choice is visible, tuning the timing of when the AI steps in, and matching Zymix's visual language down to the iPhone frame dimensions. The whole build history is traceable, so the process is honest about what was made during the hackathon.

Tech and styling. Sorted is a front-end prototype built with React + Vite, styled to match Zymix's visual language: the yellow-green gradient, the chat bubbles, the bottom navigation, so it reads as a Zymix-style Mini App. A lot of our effort went into the prompting and the product thinking, not just the coding.

Following the track brief, Sorted is built as an independent, standalone concept. It is a Zymix-style Mini App that shows how this feature would live inside Zymix, rather than integrating directly into the app.

Challenges we ran into

Most of the hard work wasn't the code. It was the thinking before the code.

Scoping the scenario. We narrowed down step by step. First: are we solving what to eat or where to go out? Then, within food: eating out, or cooking together? Each of these is a genuinely different problem with different angles, so getting specific early mattered.

The hardest design call: how bounded should the AI be? This was the real challenge. Too proactive and it's annoying. Step in too early and it's intrusive. Step in too late and the group has already given up. So when to intervene became the actual intelligence of the product, not an afterthought. Alongside that, we had to decide where the user's control sits. Sorted proposes, but it never decides for the group.

Making the AI's memory believable. We seed the demo with a "last week's chat" so judges can see that the preferences are earned, not hardcoded. Sorted extracts them from past conversations the user has given it permission to read. That's what makes the memory credible rather than magic.

What we learned & what's next

The biggest thing we learned is about the product itself. A good coordination tool isn't one that decides for you or throws a list of options at you. The deeper job is helping people get past the social awkwardness, past the "whatever" standoff where everyone's too polite to choose.

We also learned a lot about working with AI. Good vibe-coding comes down to refining your prompts, and that only works if you've genuinely thought through the underlying logic first. Once the foundation is clear, everything downstream is more organised.

What's next:

• Refine the UI further
• Hook into real Zymix Location and Wallet, so a decision flows straight into "drop the pin, split the bill"
• Close the loop on the group decision flow
• And most importantly: real user testing

Built With

• claude-code
• claude-opus-4.8
• css
• javascript
• react
• vite