Inspiration
Planning an outing for a group is a small coordination problem disguised as a restaurant search. One person cares about food, another wants somewhere creative, and someone else needs to keep spending light. A list of popular places leaves the group to do the hard work.
Unstressed gives that work to an agent, while keeping the people and their preferences visible.
What it does
Tell Unstressed who is coming and what you have in mind. Bring a saved circle, add someone new, or change a participant’s tastes in conversation. Food, activities, music, mood and spending preferences feed the same plan. Taste cards appear inside the chat, and edits rework the itinerary rather than creating a disconnected recommendation list.
The result is a one-to-three-stop outing with real venues, a group budget allocation and practical next steps. The agent asks for an unspecified date or time instead of silently picking tomorrow. Flexible windows are fitted around reported weekly opening hours; a specific start time stays fixed. Keep a stop you love, replace another, or ask for just one place. The plan stays within reach beside the conversation, with direct access on mobile.
A visual Connections view traces people to their stated preferences and then to each stop. It distinguishes Qloo-reported artist contributions, matching place tags and preferences used in the search without individual contribution evidence.
Circles let familiar groups plan again without starting from scratch. Google sign-in keeps account tastes and circles. Signed-in people search only their accepted circle connections; a private invitation must be accepted before a new account joins. Guests can add simple name-and-taste profiles. Stacked avatars and circle names keep the people visible in both the conversation and the result.
Autopilot brings a requested circle and its saved tastes together, applies requested changes, finds places, shapes the day and checks reported hours. Venue details include available official links, contact information and an enquiry draft. Download a group or personal daybook, open a prefilled Google Calendar event to save yourself, or share a link to the evolving plan. The agent does not send messages or make bookings.
How we built it
The interface uses React and Vite; the API uses Node.js and Express. Gemini interprets conversation into validated planning and participant tools. Application code manages the roster, scheduling, budget arithmetic, kept stops and revisions.
Qloo is the cultural intelligence layer. We resolve explicitly supplied artists to Qloo entities and food or activity preferences to Qloo tags. Each participant’s signals retrieve real place candidates. The planner checks category and location evidence, balances group rankings, and chooses distinct stops. Evidence from each participant’s own results supplies the Connections view.
Without Qloo, the product loses the cross-domain retrieval and provider evidence that make those recommendations and taste connections possible. The language model is not allowed to invent the venue list.
Challenges we ran into
A preference used in a search is not proof that it caused a recommendation. We had to preserve that distinction in both the data and the visual explanation. We also had to prevent one participant from inheriting another person’s evidence.
Group edits introduce concurrency: a swipe or new participant can arrive while the previous plan is still being assembled. Versioned updates prevent older results from replacing newer choices. If discovery fails, the last complete plan remains visible and is clearly marked as previous.
Accomplishments and what we learned
The central loop works with real Qloo results: change a person, their tastes or the brief, then see the shared plan respond. We have exercised live Gemini conversation, Google sign-in, single-stop outings and plans in Ghana, London, Paris, New York and Tokyo.
Cultural grounding is most useful when people can inspect it. Showing the path from a person to a preference to a place makes the agent’s reasoning easier to question and refine.
What’s next
Test the full experience with more real circles and measure how well the shared plan serves different tastes. A private owner dashboard now tracks sign-ups, active accounts and feature usage without collecting chat text or detailed tastes in its events. Next are verified route estimates and date-specific availability. Reported weekly hours are not confirmation for a particular date; budgets remain allowances, and prices, travel times and individual suitability need confirmation.
Log in or sign up for Devpost to join the conversation.