GlowGuard AI - Resilient AR Beauty & Skincare Advisor

AI-powered skin analysis + AR virtual try-on that never goes down even when the cloud does.

Comment

Inspiration

The Problem: $180B Industry, Zero Personalization at Scale:

  • Walk into any beauty store and you face 500+ products with zero guidance.
  • Ask a salesperson and you get a commission-driven recommendation. Buy online. and you're guessing. “70% of skincare purchases are wrong for the buyer's skin type and are” returned, abandoned, or simply wasted. Meanwhile, enterprise-grade AI beauty tools exist but only for brands spending $500,000+/year on proprietary tech. The average consumer and small beauty retailer? Left out entirely.

GlowGuard AI changes that.

What It Does

GlowGuard AI is an intelligent, resilient beauty advisor that combines AI-powered skin analysis, augmented reality virtual try-on, and personalized skincare recommendations, all in one seamless web experience.

The Problem We Solve

Walk into any beauty store and you face 500+ products with zero personalization. Buy online and you're guessing. “70% of skincare purchases are wrong for the buyer's skin type” returned, wasted, or simply sitting unused on a shelf.

Enterprise-grade beauty AI exists but only for brands spending $500,000+ per year. The everyday consumer and small beauty retailer are left out entirely.

GlowGuard AI changes that.

Core Features

  1. AI Skin Analysis Upload or capture a photo and our AI instantly analyzes your skin across multiple dimensions:
  2. Skin type classification: dry, oily, combination, normal, or sensitive
  3. Concern detection: acne, dark spots, wrinkles, enlarged pores, texture
  4. Skin health score: personalized 0–100 rating
  5. Priority ranking: which concerns to address first

No guesswork. No generic advice. Pure data-driven personalization.

  1. AR Virtual Try-On See makeup products on your actual face before buying in real time:
  2. Lipstick: precise lip contour detection and color application
  3. Eyeshadow: gradient blending across eye region
  4. Foundation: full face coverage matching your skin tone
  5. Blush: natural radial gradient on cheekbones

Powered by *”MediaPipe Face Mesh” with 468 facial landmarks for pixel-accurate placement across all skin tones and lighting conditions.

  1. Resilient AI Skincare Advisor Get a personalized morning and evening skincare routine generated by AI and powered by “TrueFoundry's AI Gateway”:
  2. “Primary LLM”: Claude claude-sonnet-4-20250514 (Anthropic)
  3. “Fallback 1”: GPT-4o (OpenAI)
  4. “Fallback 2”: Gemini Flash (Google)
  5. Final fallback”: Intelligent cached responses

When one AI provider goes down, another instantly takes over. “The user never sees an error. The experience never breaks.”

  1. Live Resilience Dashboard Watch the AI infrastructure in real time:
  2. Which LLM provider is currently active
  3. Provider health status (green/yellow/red)
  4. Automatic failover notifications
  5. Response time per provider

This transparency builds user trust and demonstrates enterprise-grade reliability in action.

Who It's For

“Everyday consumer”:- Know your skin type, try before you buy. “Beauty enthusiast”:- Discover products matched to your concerns. “Small beauty retailer”:-White-label AI personalization for your store. “Beauty brand”:- API integration for e-commerce product pages.

The Business Case

GlowGuard operates in the “$180 billion global beauty market” with two clear revenue paths:

“B2C (Direct to Consumer)”

  • Free tier: 3 skin analyses per month
  • Premium: $9.99/month: unlimited analyses, routine tracking, product history, and personalized ingredient alerts

“B2B (API Licensing)”

  • Skin analysis API: $0.05 per analysis
  • Virtual try-on integration: $0.10 per session
  • Target clients: beauty retailers, e-commerce platforms, salon booking apps

“Year 1 projection”: 10,000 premium users + 5 B2B clients = ~$1.5M ARR

How we built it

GlowGuard AI is a full-stack web application that delivers three core experiences:

  1. AI Skin Analysis Upload or capture a photo and our AI powered by “Perfect Corp's Skin” Analysis API: analyzes your face across 14 dimensions:
  2. Skin type classification (dry / oily / combination / normal / sensitive)
  3. Concern detection (acne, dark spots, wrinkles, pores, texture)
  4. Skin health score (0–100)

  5. Personalized concern priority ranking

  6. AR Virtual Try-On Try lipstick, foundation, and eyeshadow “before you buy” using “Perfect Corp's Virtual Try-On API”. Select any product from our catalog, see it on your actual face in real time, and make confident purchase decisions.

3.Resilient AI Skincare Advisor Get a personalized morning + evening skincare routine generated by AI with a twist: “it never fails, even when the AI infrastructure does.” Using “TrueFoundry's AI Gateway”, GlowGuard routes requests through multiple LLM providers with automatic failover

Challenges We Ran Into

Building GlowGuard AI in 5 days taught us more than we expected. Here are the real problems we hit and how we solved them.

Challenge 1: AR Virtual Try-On Placement Accuracy

The problem: Our initial AR implementation produced inconsistent facial landmark detection lipstick bleeding outside lip boundaries, eyeshadow misaligned with eye regions, and foundation not following the natural face contour. The pixel-precision requirement for beauty AR is unforgiving: even a 5-pixel offset looks obviously wrong.

What we tried: We initially integrated a third-party beauty AR API. While it handled the heavy lifting conceptually, it struggled with:

  • Diverse skin tones (darker skin tones detected less reliably)
  • Varying lighting conditions (low light = landmark drift)
  • Different face shapes and angles

How we solved it: We switched to “MediaPipe Face Mesh” Google's open-source 468-point facial landmark detection model with refineLandmarks: true for iris-level precision. This gave us:

  • 468 landmark points vs the typical 68-point detection
  • Consistent performance across all skin tones
  • Real-time 30fps tracking on average hardware
  • Full browser compatibility no backend needed
  • Zero API costs or rate limits

We then built custom Canvas API rendering for each makeup type polygon fill for lips, gradient blending for eyeshadow, radial gradients for blush all mapped precisely to MediaPipe's landmark indices.

“Lesson learned:” The best tool for the job isn't always the most obvious one. Switching to an open-source foundation gave us more control, better results, and eliminated dependency on external API quotas.

Challenge 2: LLM Response Consistency Across Providers

“The problem:” When Claude generates a skincare routine, it follows a specific conversational structure. When GPT-4o generates the same routine, the formatting differs. When Gemini responds, the structure changes again. Our frontend was receiving three completely different JSON shapes depending on which provider responded causing the UI to break unpredictably during failover.

How we solved it: We built a “response normalization layer” a Pydantic model called ‘SkincareRoutine` that every provider's output is parsed into before reaching the frontend:

Accomplishments That We're Proud Of

  1. Built a Production-Grade App in 5 Days Full deployed app real AI, real AR, real failover. Not a concept. Shipped.

  2. The Resilience Architecture Actually Works Live chaos demo trigger Claude failure, watch GPT-4o take over seamlessly. We prove it, not just describe it.

  3. 468-Point AR That Actually Tracks Correctly 30fps real-time tracking, all skin tones, runs entirely in browser, sub-millimeter lip and eye precision.

  4. Two Sponsor Challenges, One Coherent Product Perfect Corp + TrueFoundry aren't bolted-on features they're core architectural pillars.

  5. A Clear Path to Real Business $180B market + B2C freemium + B2B API = believable startup story with real revenue projections.

  6. Honest Engineering Live demo, real camera, real AI, real failures handled gracefully. That's the accomplishment we're most proud of.

What we learned

On the technical side:

  • AR try-on at consumer scale is harder than it looks the difference between "it works" and "it looks good" is enormous in beauty tech
  • Multi-LLM resilience isn't just redundancy it's an architectural discipline requiring normalization layers
  • Perfect Corp's APIs are genuinely powerful; the skin analysis depth surprised us

On the product side:

  • Users trust AI beauty recommendations more when they can see the reasoning ("your dry skin needs ceramides because...")
  • The virtual try-on is the feature people want to show friends, it's inherently shareable
  • Resilience is invisible when done right and that's exactly the point

What's next for GlowGuard AI - Resilient AR Beauty & Skincare Advisor

Immediate (Next 30 Days)

  • [ ] Add 500+ product catalog from real beauty brands
  • [ ] Implement product purchase links (affiliate revenue model)
  • [ ] Add skin progress tracking (weekly photo comparison)
  • [ ] Mobile app (React Native)

Short-Term (3 Months)

  • [ ] B2B API: white-label for beauty retailers
  • [ ] Shopify plugin for beauty e-commerce stores
  • [ ] Ingredient scanner (photograph any product, AI analyzes ingredients)

The Startup Path GlowGuard operates in a “$180B addressable market” with two clear revenue streams:

B2C: Freemium model

  • Free: 3 analyses/month
  • Premium ($9.99/month): unlimited + routine tracking + product history

B2B: API licensing

  • White-label skin analysis for beauty brands: $0.05/analysis
  • Virtual try-on integration for e-commerce: $0.10/try-on
  • Target: Sephora, Ulta, independent beauty retailers

Year 1 projection: 10,000 premium users + 5 B2B clients = ~$1.5M ARR

Built With

  • esbuild
  • express.js
  • googlefonts
  • html5canvasapi
  • lucidereact
  • motion-framer
  • node.js
  • perfectcorpapi
  • perfectcorppegasus-1000
  • react18
  • tailwindcss
  • truefoundryairesiliencegateway
  • typescript
  • vite
Share this project:

Updates