Side project

Quiet Table is a taste-driven restaurant finder for you and your mates.

Hypothesis

The future of agent-powered user experiences isn’t only a chat box. Efficiency, delight and precision come from the user and the product having shared information, context and data. I see this as a taste profile. A brand-first product will mix agent, personalisation and brand moments to enrich the UX, forming new patterns, content and UI when needed.

Goal

To learn by building: an agentic UX for a simple, common use case.

Focus

Finding a good restaurant based on your tastes, patterns and friends.

Great places to eat, especially in unfamiliar cities, can be hard to find. Where people eat says a lot about them. Like fashion and music, it’s a flex and part of their social currency. Somewhere between Michelin stars, Instagram, Beli and maps apps, I believe there’s space for a restaurant finder that adds to it whilst being genuinely helpful for others in your social circle.

Notes

Designing the agents’ layers of thinking: context, location and dining needs. The system pulls from the user’s calendar and preferences to present a next step that feels personal, yet surprising.

Stack

Claude, Astryx Design System, Figma, Clerk, V0, GitHub

Value model: personal taste, the social graph and agentic UX, overlapping Flow overview: from onboarding and context, through a specific need and rejoining the journey, to results

Research

The opportunity is real. Beli is growing fast in the US (120 million restaurant ratings). Google and Apple Maps lists are neither inherently social nor discovery-driven. Review systems are outdated and unreliable. There’s space for a fun, taste-driven foodie UX.

How people find restaurants today: blog guides, a sticky note on friend-backed recommendations, and maps apps Choosing where to eat on a phone at a shared table of food

Agent logic

How might we control the agent’s responses, and build a system that actually considers a user’s taste and preferences?

Agent Vs Wizard
Agent Vs Wizard
Social Graph and Personalisation
Social GraphPersonalisation
Architecture
Architecture
Testing search: the agent replies with restaurant suggestions as cards
Testing Search
Agent Vs Wizard
Agent Vs Wizard

UI

Meta’s Astryx is an open-source design system that’s fully customisable and agent-ready. I wanted the system to produce the right UI elements based on user requests. Designing a vertical flow is easy enough: you can tell the wizard which components to use, like time cards or calendar pickers.

When a user leaves the set vertical flow, knowing which component to surface becomes more difficult for Astryx. ChatGPT 6 recently released “Intelligent UI”, solving for exactly this pattern.

Asking the agent: does Esa like dumplings? Friends’ tastes: the restaurants each member loved or liked
Choosing who’s coming, to weight their taste in the search Places you enjoy, ranked as Loved, Liked or Fine
Casual, Agent Off
Date Night, Agent On
Personalised results: date-night picks in London matched to Molly’s taste
Personalised Results
Ask the agent: the chat input
Ask The Agent

Brand

Leading up to this project, I also wanted to think about what a high-end lifestyle app could feel like if it were to scale: what sort of look and feel builds trust, feels solid and hints at luxury.

Brand exploration: type, motion and a palette drawn from desert stone
Mood board: carved stone, wood grain and hands Early sketches of the app in a notebook
The agents at work: logistics, dining, weather and personal preferences, then planning and location
The voice and input bar Your Sunday evening is planned: the restaurant and the car home, booked

Contact