About the project
Inspiration
The global beauty market suffers from information asymmetry: customers often purchase products without confidence, leading to returns, dissatisfaction, or neglecting key skincare routines. Studies suggest up to 40% of consumers abandon products due to poor personalization or uncertainty about efficacy.
Nuvia Beauty was inspired by this gap: using AI to provide confidence-driven, explainable recommendations, integrating real-time analysis, and capturing structured beauty profiles. It addresses the underestimated retail segment where physical shops lack digital augmentation for personalized skincare guidance.
Research Insights Consumer Negligence: Many customers underuse products or mismatch skincare due to uncertainty. Dermatology research indicates that ~60% of users misapply products or discontinue prematurely, affecting both satisfaction and business retention. AI Confidence Shopping: Generative AI, skin analysis algorithms, and predictive scoring can guide consumers to products suited to their skin tone, type, and concerns. Explainable AI increases trust and repeat engagement. Underrated Market: Small- and medium-sized beauty retailers remain underserved in digital personalization solutions. AI-augmented consultation can increase conversion rates and lifetime value, offering a high-margin growth opportunity. Project Overview
Nuvia Beauty bridges AI intelligence with in-store experiences:
Seller-Assisted Consultation: Capture or upload customer photos via PWA or tablet interface. AI-Powered Analysis: Use Perfect Corp APIs for skin tone, undertone, and concern analysis. Structured Profile Storage: Securely store snapshots in private S3-compatible buckets. Deterministic Recommendations: Generate product suggestions with explainable scores and warnings. Profile Reuse: Enable returning customers to reopen profiles without rescanning.
Future expansions include:
Customer Self-Scan & History: Empower consumers to update their profiles from home, compare snapshots over time, and regenerate recommendations. Try-On Studio & VTO: Simulate makeup application virtually to increase confidence before purchase. Personalized Domains: Shop-specific branding and reusable consultation links. Enhanced AI Scoring: Incorporate vector search and multi-provider AI for advanced personalization. Gamification & Engagement: Reward adherence to skincare routines with progress tracking and gamified incentives. Beauty shopping has a confidence problem. Customers are not only asking, “Does this product look good?” They are asking whether it fits their skin tone, undertone, skin type, visible concerns, preferences, and past product response. In physical shops, sellers often give useful advice, but that advice usually disappears after the conversation.
Nuvia Beauty was inspired by that gap: the need for a beauty consultation system that works for real shops, not just online filters. The goal is to help sellers capture or upload a customer photo, run AI-powered analysis, create a reusable beauty profile, and recommend products with clear reasons instead of vague suggestions. The project is designed as a retail confidence platform, not a medical diagnostic tool. The proposal defines the product goal as increasing pre-shopping confidence by using Perfect Corp/YouCam APIs, structured beauty profiles, explainable product guidance, and product-effect history.
What it does
Nuvia Beauty supports a seller-assisted beauty consultation flow:
A shopkeeper opens the Beauty Kiosk PWA on a tablet. The customer photo is captured or uploaded with guided quality checks. The backend securely processes the image through a Perfect Corp skin or tone analysis flow. The system creates a structured beauty profile snapshot. Products are scored using deterministic rules based on skin concerns, tone, undertone, preferences, and avoid tags. The seller sees recommendation cards explaining why each product is a strong or weak match. The customer can later reopen the saved profile and recommendations without needing to retake a photo.
The strongest demo flow is seller-assisted: open kiosk, capture photo, run analysis, create profile snapshot, show recommendation card, save the profile, and let the customer reopen it later.
How we built it
We built Nuvia Beauty as a PWA-first beauty module beside the existing Zyro Fashion commerce system. The project uses the existing commerce, authentication, product catalogue, seller account, and admin infrastructure, while adding a separate beauty experience with customer, seller/kiosk, and admin surfaces.
The technical architecture is built around:
Laravel backend for beauty domain APIs, storage handling, quota enforcement, AI task orchestration, profile persistence, and recommendation logic. Next.js + React PWA frontend for customer and seller/kiosk experiences. Perfect Corp / YouCam APIs for the MVP AI skin or tone analysis. MediaPipe / browser camera validation for capture quality before spending API quota. MinIO / S3-compatible storage for private photo and result storage. Deterministic recommendation scoring before adding advanced vector or ranking models.
The proposal sets the main direction clearly: PWA-first, Perfect Corp APIs for MVP, self-hosted MinIO for local/staging private storage, and deterministic rules + scoring before vector/ranking systems.
What we learned
We learned that the hardest part of beauty AI is not only running an AI model. The harder problem is connecting analysis results to a safe, useful, and explainable shopping journey.
Key lessons:
Explainability matters more than magic. A recommendation is more useful when it says why a product matches the user’s profile. Privacy must be designed early. Raw photos should not become the permanent identity of the customer. The structured profile snapshot should be the reusable record. Backend-only AI orchestration is safer. Perfect Corp keys, MinIO credentials, and signed URLs must not leak to the frontend. Seller workflow is the strongest MVP. A physical-shop-first kiosk flow creates clearer retail value than starting with full customer self-scan. Scope control matters. Makeup VTO, customer self-scan, multi-provider AI, analytics, and vector search are useful later, but they would weaken the MVP if added too early.
The MVP boundary intentionally excludes full customer self-scan, offline encrypted drafts, full Try-On Studio, multiple Perfect Corp APIs, full analytics, billing, cross-profile prediction, pgvector, native mobile, social feed, and clinical diagnosis.
Challenges we faced
The biggest challenge was keeping the project realistic while still making it impressive for a hackathon.
We had to balance:
AI capability vs. demo reliability: live AI is powerful, but provider outages or quota limits can break demos. So the system supports demo fallback. Beauty personalization vs. medical safety: the system must use retail-safe language like “profile-based confidence score” and avoid claims like “diagnoses your skin condition” or “clinically proven to treat.” Privacy vs. personalization: storing raw images forever would be risky, so the design favors temporary encrypted media and reusable structured profile snapshots. Scope vs. ambition: the project has many possible extensions, including makeup try-on, hair, jewelry, analytics, and vector search, but the MVP focuses on one strong loop. Frontend simplicity vs. backend safety: the frontend must stay simple while the backend handles secure provider calls, storage, quotas, task lifecycle, and audit logs. What makes it different
Nuvia Beauty is not just a beauty filter. It is a beauty intelligence and recommendation system for real shops.
The difference is the loop:
Seller consultation → AI analysis → structured profile → explainable recommendations → saved history → better future guidance
Instead of forcing the customer to scan again every time, the system moves toward a reusable customer beauty profile. That makes the experience faster, safer, and more valuable over time.
Log in or sign up for Devpost to join the conversation.