The Challenge

FaceCake sought to enhance the online shopping experience and improve product discovery while boosting sales by increasing both conversions and total items purchased per transaction. Additionally, they aimed to remove barriers to conversion by providing a personalized, interactive shopping journey where questions are answered seamlessly via a next-generation chatbot, increasing sales while reducing returns and the number of calls to call centers.

“FaceCake has developed a range of Computer Vision and Machine Learning solutions focused on face and body tracking, as well as recommendation engines. We discussed our AI initiatives with Google as part of our partner program, aiming to deploy them on Google Cloud. One initiative stood out: an AI Personal Stylist incorporating Google’s LLM, integrated with FaceCake’s AI and AR platform and deployed on e-commerce sites. We wanted to leverage Google’s Gen AI as the top layer of our tech stack. Following Google’s suggestion, we started collaborating closely with Wursta,” explained Ashwin Rajendraprasad, Chief AI Officer at FaceCake.

The Solution

FaceCake engaged Wursta, an AI & digital workplace consultancy, to quickly build a generative chat interface for AVA (Acquire, Validate, Adapt), its cutting-edge AI and AR personal shopper and stylist that engages digital consumers in a unique way. “The Wursta team quickly understood what we wanted to launch and hit the ground running,” said Ashwin.

The Results

AVA is the first of its kind to transcend traditional search methods and empower users with proactive product discovery. AVA initiates interaction with shoppers, as opposed to conventional chat UIs where users typically start a conversation. Through personalized recommendations based on users’ activity, both traditional and augmented, AVA enables shoppers to virtually ‘Try-On’ products enhancing the product discovery process.

Personalized Recommendations & Virtual Try-On

With a multimodal interface, AVA is capable of processing a variety of informational inputs including images, videos, and text. Once a user grants permission for AVA to access their camera, the AI integrates visual input such as the user’s face shape, skin tone, eye color, and expression (sentiment). Powered by FaceCake’s novel AI, AVA offers personalized recommendations tailored to the user’s preferences. With FaceCake’s expertise in visual data, AVA provides real-time feedback and custom advice to users, continuing throughout the Try-On process.

“AVA is extremely unique. Unlike any others in the market, AVA combines AI, AR, and an Auto-Regressive LLM,” added Ashwin.

“The feedback has been amazing. Users really appreciate how easy it is to find products. You’re chatting with a personal stylist, and you can one-stop-shop. Unlike traditional LLMs, AVA engages the customer in back-and-forth discussions. So many sites overwhelm you with options, but AVA enhances the product discovery process, making it easier for shoppers to quickly find what they want. For people who aren’t looking for something specific, AVA serves as an in-store personal assistant and can make suggestions,” said Ashwin.

Users Can View Themselves “Wearing” Products with FaceCake’s Augmented Reality (AR)

Leveraging FaceCake’s robust AR Try-On platform, AVA allows shoppers to virtually Try-On products during a live session.

AVA integrates with Google’s next-generation model: Gemini 1.5 Pro. Using the LLM, AVA can engage users in a conversational, guided shopping experience and provide them with FaceCake’s product recommendations while showing themselves “wearing” products with a real-time camera feed.

Autoscaling & Support for Future MLOps and Training Pipeline Needs

The collaboration leverages Google Cloud infrastructure, ensuring the scalability and responsiveness best suited for real-time applications. AVA is deployed through Vertex AI to allow for inference autoscaling, as well as to support future MLOps and training pipeline needs.

“Many of our APIs were running on our secure cloud and servers. Wursta helped us migrate the infrastructure and deploy AVA on Google Cloud. We can now deploy and scale the solution easily to multiple e-commerce partners, including top fashion and beauty brands,” said Ashwin.

Easily Tailored for Consistent Branding

Additionally, AVA can function independently or be embedded on any e-commerce site. Brands can easily tailor the experience and branding to be consistent with their products and aesthetics. “It’s very exciting to see FaceCake’s AVA bring to market exactly what e-commerce was meant to be with AVA’s immersive blend of interactive AI and AR,” said FaceCake’s Founder and CEO, Linda Smith. “Our collaboration with Google Cloud and Wursta played a pivotal role in its realization.”

“All the Wursta engineers were great; extremely helpful, supportive, and responsive,” concluded Ashwin.