DermaLens

DermaLens

Inspiration

Skincare is deeply personal, yet most people rely on trends, trial and error, or generic advice. Millions deal with acne, redness, oiliness, or irritation without access to objective, data driven guidance. Dermatologist visits are expensive and hard to schedule, and online quizzes ignore what your skin actually looks like.

We set out to build something different: an intelligent system that sees your skin, scores it objectively, and generates a routine grounded in dermatological rules, not marketing. DermaLens is designed to evolve alongside the user's skin over time.

What It Does

DermaLens is an AI powered skin analysis platform that generates personalized skincare routines backed by measurable progress tracking.

Users upload three facial photos (front, left, and right angles). Gemini Vision evaluates five key indicators: acne severity, redness, oiliness, dryness, and texture, assigning structured scores on a 0 to 100 scale. Users also specify their biggest concern and sensitivity level to ensure personalization goes beyond image analysis alone.

The system then generates a morning, evening, and weekly skincare routine tailored to the user's unique profile. To ensure stability and safety, routines are locked for a minimum of two weeks, aligned with the skin's natural turnover cycle. Weekly scans compare scores against previous results, and adjustments are made only when improvement stalls or conditions worsen.

An integrated AI chat provides context-aware skincare guidance, drawing from the user's latest scan, active routine, and stated concerns, without ever diagnosing or prescribing.

How We Built It

DermaLens follows a modular full stack architecture with clear separation between AI analysis and deterministic optimization.

Gemini 2.5 Flash analyzes facial images and outputs structured skin metrics via multimodal vision
A FastAPI backend (stateless, S3 backed) converts metrics into a structured skin profile with zero database overhead
A rule based routine engine ranks ingredients using weighted optimization based on user priority, skin conditions, ingredient effectiveness, and irritation risk
A conflict detection module prevents unsafe ingredient combinations (e.g., retinoid + strong acid)
A scheduling system adjusts application frequency based on sensitivity level
A lock policy prevents over adjustment during short term fluctuations
A SwiftUI frontend delivers a polished native iOS experience with a guided 5 step workflow
Claude Code accelerated development as an AI pair programming assistant across the full stack

By separating AI analysis from deterministic optimization logic, we ensured that every recommendation is explainable, consistent, and safe.

Challenges We Ran Into

Extracting structured, reliable outputs from multimodal AI models and validating image quality before analysis
Designing a safe ingredient conflict resolution system that accounts for real dermatological interactions
Preventing routine over adjustment during short term score fluctuations versus genuine trend changes
Aligning adaptation logic with biological skin turnover cycles of 2 to 4 weeks
Integrating iOS, backend, AI services, and AWS cloud infrastructure within hackathon time constraints
Deploying to EC2 and debugging networking issues (subnets, security groups, environment variables) under time pressure

Accomplishments We're Proud Of

Built a fully functional end to end system from photo upload to AI analysis to personalized routine generation to AI chat in a single hackathon
Created a weighted optimization algorithm for ingredient selection rather than relying solely on generative AI
Implemented adaptive logic that compares weekly scores before modifying any routine
Shipped a zero database architecture with all persistence backed by S3 JSON files
Delivered a polished native iOS interface with animated score rings, timeline style routines, and context-aware AI chat

What We Learned

Combining AI with deterministic systems creates stronger and safer health related applications. While generative AI is powerful, applications involving personal care demand structure, explainability, and constraint based logic. You cannot let a language model freestyle someone's skincare routine.

We also gained deep experience integrating multimodal AI (Gemini Vision), designing adaptive optimization engines, building full stack applications connecting SwiftUI to FastAPI to AWS, and using AI assisted development tools to ship faster without sacrificing code quality.

What's Next for DermaLens

DermaLens Premium: Unlimited AI chat access via paid subscription with a free tier of 3 messages per day
Brand Partnerships: Curated, science backed product recommendations with in app purchase links from partnered brands
Haptic feedback, achievement sounds, weekly streaks, and milestone badges to reward consistency
Push notifications for routine reminders and weekly scan prompts
Long-term progress visualization with side by side photo comparisons and trend analytics
Android client

Our long term vision is to make intelligent, adaptive skincare accessible to everyone and build a sustainable business around it.