GlowScan

Scan your face. Understand your skin. Glow smarter — your skincare expert in one scan, powered by AI that gives you personalized routines, product guidance, and daily skincare advice.

Comment

Inspiration

Skincare should not be a guessing game. Yet millions of people wake up every day not knowing their skin type, following random advice from social media, and wasting money on products that were never right for them. Dermatologist visits are expensive and not always accessible — especially across Africa where quality skincare guidance is hard to find.

We built GlowScan because we believe everyone deserves a smart, personalized skincare coach in their pocket — regardless of budget, location, or prior skincare knowledge. One scan. Real answers. Zero guesswork.

What it does

GlowScan is an AI-powered skincare coach that analyzes your face and delivers a complete, personalized skincare plan instantly. Here is how it works:

You take a selfie or upload a photo, GlowScan first validates that a real human face is present before any analysis begins You enter your gender and age range Our AI analyzes your skin tone using the Fitzpatrick scale, identifies your skin type, and detects visible concerns like acne, dark spots, enlarged pores, and dryness — based on what is actually visible in your photo, not a generic template You receive a complete personalized plan including your morning and evening routine, the right product types and key ingredients for your skin, what ingredients and habits to avoid, lifestyle advice on water intake, diet, sun exposure and sleep, and gender-specific tips only where biology actually changes the recommendation If severe concerns are detected, GlowScan automatically flags a dermatologist referral

GlowScan is a skincare coach, not a medical tool. Every result includes a clear disclaimer and speaks to you like a knowledgeable friend.

How we built it

Frontend: HTML, CSS, JavaScript with a warm beauty-focused design system built from scratch — rose, cream, gold, and dark brown palette with Playfair Display and DM Sans fonts Backend: Node.js handling image processing, API calls, and result delivery AI Engine: Anthropic Claude API with a carefully engineered system prompt that returns structured JSON only — no prose, no hallucination, no generic responses Image Processing: Canvas API compresses every photo to 800px width at JPEG 70% quality before sending to the API — cutting payload size by up to 40x and keeping response time under 4 seconds

Challenges we ran into

Getting the AI to be specific, not generic Early versions returned the same templated advice to every user regardless of what was actually in their photo. We solved this by rewriting the prompt to force specificity — every concern must include its location on the face, every product recommendation must reference the actual detected skin type, every routine step must be justified by what was seen. Face validation without a dedicated vision API We needed to reject random images food, landscapes, screenshots without using a separate face detection API. We solved this by building face validation directly into the AI prompt as the first mandatory step, returning a structured error JSON that the frontend handles gracefully with a friendly retry card. Camera access silently failing Live camera capture worked in upload mode but failed silently in camera mode. The root cause was mixed HTTP/HTTPS contexts, camera access requires a secure origin. We fixed this with an explicit protocol check, specific error messages for every failure type (NotAllowedError, NotFoundError, NotReadableError), and a clean retry flow. Using gender without creating bias Adding gender to skincare sounds simple but done wrong it becomes stereotyping. We spent significant time defining exactly where biology changes skincare science sebum production thickness, shaving irritation, hormonal acne patterns versus where it does not. Gender now activates in only 4 specific biological contexts in our entire system. Making it fast on mobile Raw phone photos sent to the API were 3–5MB and caused slow responses. Compressing to 800px JPEG 70% brought that down to around 80KB — a 40x reduction — without visible quality loss for skin analysis.

Accomplishments that we're proud of

Built a complete working AI skincare analysis system in hackathon time Engineered a prompt that returns specific, face-accurate advice not generic templates every single time Designed a UI from scratch that looks and feels like a real beauty brand, not a prototype Implemented responsible AI features: face validation, dermatologist escalation flag, and a non-negotiable disclaimer on every result Used gender personalization the right way grounded in biology, never in stereotypes Achieved sub-4-second response times on mobile through image compression and token capping Built a dermatologist directory with real Kigali hospitals so users always have a path to professional care when they need it

What we learned

Prompt engineering is product design. Every word in a system prompt is a decision that directly affects what the user sees. Vague instructions produce vague results. Image compression is the highest-impact performance optimization for AI image apps — faster than any other single change we made. Responsible AI means knowing when to stop. The face validation step and the dermatologist flag taught us that a good AI feature is as much about what the system refuses to do as what it does. Gender personalization requires research, not assumptions. We read actual dermatology literature before writing a single line of gender-related prompt logic. Warm tone matters more than technical accuracy when users are anxious about their skin. A kind, clear response builds more trust than a technically perfect cold one. Real data beats empty states every time. Showing a populated dashboard with a previous scan score and skin type communicates a working product in two seconds to any judge or user.

What's next for GlowScan

Live Dermatologist Connect : in-app booking for real consultations via chat or video with licensed dermatologists across Rwanda and beyond Weekly Progress Reports : automated skin improvement scores comparing scans over time with visual trend charts Product Ingredient Scanner : scan a product barcode to instantly check whether its ingredients are right or harmful for your specific skin Expand the dermatologist directory : partner with clinics across East Africa to make professional skin care accessible in every country Routine reminders : gentle daily nudges to help users stay consistent with their personalized morning and evening routines

Built With

Share this project:

Updates

Submission history