Back to all work
AI PRODUCT DESIGN • MOBILE APP CONCEPT • BEHAVIORAL DESIGN
Tiny Life Experiments
Designing an AI-powered app that helps users design structured 7-day behavior experiments that are small and insight generating. Tiny Life is not a habit-tracking app, but a behavioral experimentation tool to help users learn and grow.
*This is a fully-functional working prototype that I developed using Claude Code and Google AI Studio.
ROLE
Product Designer & AI Architect
TYPE
Concept | Side Project
When I committed to a daily creative practice, I expected the hardest part would be discipline. In fact, it was direction and growth. When I searched for inspiration, I mostly found generic drawing prompts that weren’t connected to the concepts or medium I was exploring. When I felt stuck on a concept, I wanted feedback to challenge me and help me evolve. Although my journals quickly filled with daily sketches, there wasn’t the sense of progression that I was hoping for.
Atelier is my attempt to build the kind of creative companion I was looking for. One that could respond to my work, track patterns, challenge me in ways that felt personal, and evolve alongside me as an artist.
CHALLENGE
Atelier's AI engine operates across three distinct modes. Each mode has its own output structure, tone calibration, and design constraints. Together they create a practice loop: the prompt drives creation, the critique generates learning, and the insights reveal the larger arc of an artist's development over time.
CORE MODES
Three Modes of Intelligence
1
Daily Prompt Generation
Crafted, intentional prompts that match the artist's skill level, medium preferences, time constraints, and introduce friction gradually. Prompts include a creative challenge, skill emphasis, a constraint, an optional stretch variation, and references to relevant artists or artwork.
Difficulty matched to performance
No repeats of recent prompts
Intentional creative constraints
2
Artwork Feedback
Structured, medium-aware critique covering composition, light and value, color or tonal range, and technical execution specific to the artist's chosen medium. Every review closes with exactly one refinement suggestion.
Medium-specific criteria
Never more than one improvement
Growth signals vs. past work only
3
Longitudinal Insights
Weekly and monthly reflections that surface emerging strengths, recurring patterns, and suggested focus areas. The system compares the artist only to their own past self, tracking trajectory rather than benchmarking against external standards.
Emerging strengths (2-4)
Patterns to explore (1-3)
Focus recommendations
Challenge That Grows With You
The prompt engine doesn't just randomize difficulty. It adjusts dynamically based on the artist's recent performance, using critique scores to determine whether to push harder, maintain the current level, or simplify. The algorithm evaluates three distinct signals: recent performance trends, pattern saturation in the artist's work, and which axis of complexity to increase next.
Step 2
Detect Pattern Saturation
If the artist repeatedly centers compositions, uses similar palettes, avoids complex anatomy, or skips backgrounds, the system introduces targeted constraints to break the pattern.
Step 3
Adapt on One Axis Only
Difficulty increases via technical constraint, conceptual depth, or time compression. Only one axis at a time, preventing the overwhelm that comes from raising everything at once.
ADAPTIVE DIFFICULTY
Step 1
Evaluate Recent Performance
Average scores over the last five entries determine direction. Above 85 increases difficulty. Between 70 and 84 maintains level. Below 70 reduces complexity, with significant simplification below 55.
Invisible Intelligence
Behind every piece of feedback is a structured scoring rubric. Each artwork is internally evaluated across five categories, each scored 0 to 20 for a total of 0 to 100. These scores are never displayed to the user unless explicitly requested. They exist purely to power the adaptive engine: determining prompt difficulty, calibrating feedback depth, and detecting growth over time.
SCORING RUBRIC
Critique That Knows Your Tools
A watercolor painting and a charcoal drawing require fundamentally different feedback. Atelier adjusts its critique vocabulary based on the artist's selected medium. Pencil and charcoal work is evaluated on line confidence, pressure variation, and value depth. Ink work focuses on decisiveness, negative space, and bold contrast. Watercolor considers wash control, edge softness, and pigment layering. Oil and acrylic critique addresses texture, blending transitions, and layer build. Digital work is assessed on brush consistency, lighting realism, and edge control.
Ink critique card: line confidence, saturation, contrast
MEDIUM AWARENESS
Watercolor critique card: wash control, edge softness, pigment layering
Charcoal critique card: line confidence, pressure, value depth
The Voice of a Quiet Studio
Atelier's AI voice was designed as deliberately as its UI. Calm, modern, minimal, sophisticated, thoughtful. Encouraging without exaggeration. Direct but never harsh. Every output is written in short, clean paragraphs formatted for premium UI cards, not walls of text.
The system never shames, never compares the artist to others, and never provides more than one major improvement at a time. It treats feedback as a design problem: what is the single most useful thing to say right now?
TONE & VOICE DESIGN
Do
Short, clean paragraphs
Declarative phrasing
Medium-specific observations
Quiet confidence
One improvement at a time
Periods.
Don’t
Dense blocks of text
Hedging or filler words
Generic feedback
Motivational hype
Overwhelming critique lists
Exclamation points
Level-Aware Feedback Ratios
The balance between encouragement and refinement shifts based on the artist's skill level. A beginner receives 70% encouragement and 30% refinement. An intermediate artist gets a balanced 50/50 split. An advanced artist receives 30% encouragement and 70% nuance and precision. This isn't a niceness slider. It reflects a genuine design insight: beginners need to keep showing up, while advanced artists need to be challenged precisely where their work is already strong.
Behavior Design Intent
Atelier's deepest design layer isn't functional. It's behavioral. The app is built to reinforce one identity: "I show up to the studio daily." Everything in the experience, from streak-aware prompts to reflective entry flows to growth signals that only compare against the user's own past, is designed to support that identity without gamifying it. The system encourages reflection, rewards effort, highlights progress subtly, and avoids external comparison entirely.
FEEDBACK CALIBRATION
Atelier's brand system was designed with the same restraint as the product itself. Every visual decision reinforces the core identity: a quiet space, a skilled mentor, a record of growth, a daily ritual. The palette is deliberately limited to warm neutrals that feel like natural light on a clean studio surface. Accent color appears only for functional emphasis. Typography uses ultralight weights that carry the brand's spacious character, with heavier weights reserved for hierarchy rather than decoration.
Logo System: Primary logos and wordmarks
Mission
Atelier exists to make daily art practice feel intentional, rewarding, and personal. It is a quiet studio companion that helps visual artists build consistency, receive thoughtful feedback, and see their own growth over time.
Core Identity
A quiet space. A skilled mentor. A record of growth. A daily ritual. Atelier should feel like natural light falling across a clean studio - spacious, precise, and crafted with care.
Design Philosophy
Natural light . White space . Precision . Craft
Color Palette: The Atelier palette is deliberately restrained. A monochromatic foundation of warm neutrals creates the studio atmosphere. Accent color is used sparingly for functional emphasis only.
BRAND SYSTEM
Designing the Studio Identity
Designing Atelier required more restraint than most product work. The temptation with an AI creative tool is to make it do more: more feedback categories, more detailed scores, more prompts, more insights. The hardest design decisions were about what to leave out. One improvement per critique. Never displaying scores. No exclamation points. No comparison to other artists. These constraints made the experience feel like a studio instead of a classroom.
The medium-sensitivity system was the most technically interesting challenge. Getting an AI to evaluate a watercolor wash differently from a charcoal value study required designing a branching critique architecture that most creative apps don't attempt. But it's the difference between feedback that feels generic and feedback that feels like it actually looked at your work.
The feedback calibration ratios were a small decision with outsized impact. A beginner who receives mostly precision-focused critique will stop using the app. An advanced artist who receives mostly praise will stop trusting it. Getting the balance right at each level is what makes the mentorship feel real.
WHAT I LEARNED