01Build
Trained and optimized a YOLOv11 acne model, converting it to TensorFlow Lite so analysis runs fully on-device. Private, fast, offline.
← Back to all work
SkinNavi
AI facial skin-analysis that replaces skincare trial-and-error with precision computer vision, validated with real users in the field before a line of production code shipped.
Long story short
I built and validated an on-device skin-analysis product end-to-end, from the CV model to a compliant launch.
SkinNavi analyses a user's skin from a single photo and returns actionable, research-backed skincare recommendations. I owned the whole stack: the acne-detection model, the recommendation engine, auth and user-data flows. Just as importantly, I got out of the building to make sure people actually wanted it.
02Validate
Interviewed 70 people in the field and ran live booth demos to pressure-test demand and the analysis flow with real skin, not a dataset.
03Comply
When SaMD medical-device rules surfaced, I re-platformed the MVP onto LINE LIFF with zero rework of core logic, keeping launch on schedule.
Market validation
I talked to 70 people before trusting the product.
Rather than assume, I ran field interviews with 70 people and set up a booth to demo the analysis live. That's where I learned what a "recommendation" needs to feel like to be trusted, and which parts of the flow to cut. The booth doubled as an acquisition channel: onboarding real users on the spot.
How it works
On-device intelligence, compliant delivery.
YOLOv11 → TensorFlow Lite so acne analysis runs on the phone. No image ever leaves the device.
Engineered a bespoke verification flow to turn LINE LIFF tokens into trusted app sessions.
Moved the MVP to LINE LIFF for medical-device compliance, reusing core logic to protect the timeline.
By the numbers
164users onboarded
70field interviews
3field deployments
100%on-device analysis
Roadmap
Where SkinNavi goes next.
Acne analysis & recommendations live on LINE LIFF
On-device inference, private and offline
164 users onboarded across 3 deployments
Expand analysis beyond acne: redness, texture, pigmentation
Progress tracking to show skin change over time
Tighter recommendation matching from user feedback
Curated product marketplace & brand partnerships
Full SaMD certification pathway
Regional scale beyond initial launch market