OUR TECHNOLOGY:
Envision a future where ‘AVA’, your AI-powered beauty assistant, resides on your smartphone, armed with an extensive understanding of your skin health and lifestyle. AVA transcends traditional limitations, offering unbiased, brand-agnostic advice with a comprehensive grasp of the global skincare product landscape.
AVA isn’t just a consultant; she’s a daily companion. Linked to your IoT devices, she gauges your sleep patterns, stress levels, and even adapts to your travel schedule, factoring in the environmental conditions of your current locale, like the pollution index.
POWERED BY AI
With her Large Language Model interface, AVA becomes more than an advisor; she’s a conversationalist, a savant in the realm of beauty, who understands your lifestyle intimately. This daily dialogue ensures that AVA’s recommendations are as dynamic and evolving as your life itself.
As we step into the era of Beauty 3.0, we find ourselves at a crossroads where technology intimately intertwines with personal care. AI is not merely an enhancement; it’s a transformative force set to redefine the skincare industry. Its promise lies in offering a more inclusive, effective, and deeply personal approach to beauty.
Welcome to FaceFlow AI, where technology isn’t just a tool but your most trusted beauty advisor, understanding you, your environment, and your unique needs in profound ways.
Our AI skincare expert, based on the in-house LLM, is currently in its third iteration. We have continuously updated our unique fine-tuning process to improve the model’s performance in the skincare domain. The current iteration was trained on 15 years of open-access journal articles on cosmetic dermatology. Its knowledge base also included hundreds of popular SKUs from the leading skincare brands, covering most of the trail product lines.
To evaluate the performance, we have created two separate datasets: SkinProdEval, which assesses the model’s performance on product information and recommendation, and DermSciEval, which tests the model’s knowledge of scientific analysis in cosmetic dermatology. Both metrics are based on BERTScore to measure the semantic distance between the generated answer and the reference. Our model achieves +10% and +15% performance gain over GPT 4o on these two tasks. It’s also able to run on a T4 GPU, representing >60% cost savings for businesses over GPT 4o.