Building a Science-First Ingredient Interpreter
TL;DR:
Translating Skincare Labels into Science-Backed Clarity
AI UX meets ingredient literacy, no fear, no fluff, just facts.
I’m leading UX and content design for an AI-powered skincare interpreter, built in collaboration with AI engineer Keon Sadatian. Together, we’re rethinking how ingredient data is explained, bridging the gap between raw chemical listings and everyday consumer understanding.
What’s New
Structured a skincare AI assistant using tagged ingredient data and strict, no-guess prompt rules.
Tested with 50 real-world queries to validate accuracy, logic handling, and tone compliance.
Achieved 86% accuracy and 96% strict compliance, with clear fallback behavior on missing or risky data.
Failures exposed edge cases like missing fields, unclear logic filters, and tone drift—now feeding next dataset upgrade.
The Why:
The Problem
Ingredient lists are dense, inconsistent, and often wrapped in jargon or anxiety-inducing scores. Most tools offer little context and even less trust.
Our Solution
We’re building a system that turns any ingredient list into a clear, evidence-based breakdown, grounded in formulation science, readable by real humans, and structured for scalable GenAI.
Paste a product’s ingredient list → Get clear, contextual, evidence-based explanations.
We’re Building This Tool to…
Explaining each ingredient’s function in readable, accurate language
Providing context such as concentration, formulation type, and common usage
Avoiding moralizing or fear-based terminology
Offering credible references for deeper learning and transparency
Summarizing overall product composition (e.g., “4 humectants, 2 emollients”) for quick understanding
We don’t follow “clean beauty” scoring systems like EWG. Our approach is rooted in scientific evidence—not anxiety.
What We’re Exploring:
Ingredient Data Sources
We pull from trusted sources (INCI Decoder, CosIng) and emphasize reference-backed explanations over moralizing or marketing trends. Our aim? Insight, not influence.
UX Patterns
Ingredient-by-ingredient breakdowns for clarity and education
Summary views that highlight overall product structure (e.g., ingredient roles and distribution)
Reference links to support transparency, trust, and deeper exploration
AI Capabilities
GPT models decode INCI terms into calming, accurate language.
Contextual prompts surface safety thresholds and typical use cases.
Ingredient roles (e.g., humectant, exfoliant) and actions (e.g., hydration, brightening) are surfaced clearly.
Example Output
“Phenoxyethanol: A preservative that stops bacteria in water-based formulas. Usually well-tolerated, but may cause irritation at high levels (>1%). Often found in cleansers and moisturizers.”
See: PubChem, CIR Report
Collaboration and Role:
I lead UX flows, prompt design, and content system logic.
Keon builds the frontend/backend prototype using a local Ollama runtime, enabling tight feedback loops and privacy-safe iteration.
We co-design surfacing logic and trust cues, making sure every insight feels both grounded and empowering.
Phases:
Phase 1: Testing + Iteration (Current Phase)
Integrate with LLM or retrieval-augmented model
Conduct usability testing on language clarity and risk phrasing
Align outputs with accessibility and enterprise design systems
Phase 2: Testing + Iteration
Begin user testing with real ingredient lists
Test upload and paste workflows
Build a trust-feedback loop to guide future improvements
Future Plans
Layer in user profiles (e.g., acne-prone, sensitive skin, rosacea)
Enable “smart” questions (e.g., “Is this pregnancy safe?”)
Develop a Chrome extension for ingredient popovers while brows