Empowered by AI technologies, Mira wants to create a more enriching buying experience that combines millions of consumer product reviews, side-by-side comparisons, and click-through purchasing into a single unified interface. For the moment, Mira is centered around shopping for beauty products — giving users a means of asking advice from other consumers who have similar skin types, skin tones, and aesthetic preferences.
But the beauty industry isn’t the story; what’s compelling is the process and the technology underneath it all that has broader applications to other markets and verticals. “While we’re laser focused on serving the beauty community today, our technology clearly extends well beyond,” Jay Hack, Mira CEO and cofounder, told VentureBeat in an email interview. “We can do the legwork for the consumer, powering their discovery and research with the latest and greatest in AI — this really applies in any category where the journey towards purchase is more complicated than a one-click buy button.”
He described the beauty market as a “beachhead” for Mira, calling it an industry that is “highly fragmented and underserved, the products are complex, and personal identity and idiosyncrasies play a large role in finding the things that you like.”
How it works
The first step in the process is creating a knowledge base of information and a means to consume it in a simple way. The Mira team leverages its big data and AI chops to scrape reviews from all corners of the web — from publicly available articles, blogs, and videos culled from both large publications and small sites. It even grabs key information like ingredient lists. Then, using natural language processing to create summaries, it funnels all the data into one digestible, and ideally streamlined, place: a search engine with a simple user interface that looks like a Google search page.
There’s also a community aspect to Mira; with enough users, ideally there will be significant overlap in what people on the platform are looking for. That includes aesthetic preferences, but also factors like trying to find the right foundation for a certain skin tone or a product that’s friendly to sensitive skin. Mira envisions letting users ask each other for advice and product recommendations.
With a sufficient amount of salient information on hand and a community of users available, the buyer inserts themselves into the process. Using the Mira website or app, you can search for products and reviews, but there’s also an onboarding process of sorts in which Mira asks you for a selfie. It functions as a profile photo for your account, but more importantly, Mira uses that image as the basis for building content recommendations for you using computer vision, and connecting you indirectly to others who may be looking for similar products.
Rather than build an entire model from scratch, Mira employs transfer learning to fine-tune models that have already been built and trained on publicly available datasets. “We use open source tools to do all of our in-house development and have sourced various proprietary datasets from our existing users and internal data collection tools,” said Hack. The user-uploaded images that Mira collects are part of that fine-tuning process.
The magic is in the combining of NLP and computer vision to produce a rich shopping experience that minimizes the normal friction of the process. “While we use natural language processing to build powerful summaries of textual reviews, computer vision allows us to automatically detect products in online videos, provide skintone-based filtering mechanisms and even surface high-quality images showing product applied to skin,” said Hack.
Addressing the problems of people’s faces and computer vision
Combining people’s faces and computer vision is inherently problematic. And when you get into nuances like skin tone and type, there’s a further danger of running into well-established problems like the fact that much facial recognition technology performs poorly on people who have darker skin. Mira’s approach also involves some level of storing images of people’s faces, which raises privacy concerns. Hack appears to be well aware of all of the above.
“Mira gathers user data purely on an opt-in basis and keeps all personal identifying information private, in accordance with industry best practices for consumer data protection,” assured Hack. You do have to manually take and upload your own selfie. “Legally speaking, you own all of your own content and provide us with a license to use it in our community,” he said.
He further explained that Mira guarantees that users can request to scrub their data, and that personally identifiable information “will never be shared with outside entities, except on an explicit opt-in basis.”
Hack also acknowledges the troubling performance of computer vision on certain people groups. “This is important to acknowledge and address head-on, especially for a platform like ours,” he said.
Though he didn’t get into specifics, Hack said that Mira is tapping into the growing body of work that “demonstrates how to identify and mitigate” algorithms that result in “disparate impact” — a term describing unintentional discrimination that traces its etymology to a 1971 Supreme Court case. “This includes taking measures to ensure proper representation in the algorithms’ underlying datasets, stress testing their performance, and, eventually, even talking directly to users to proactively address any concerns they may have about their in-app experience — all of which are core tenets of Mira’s engineering culture,” said Hack.
There’s not really a limitation to the markets and verticals Mira can port its technology to, as long as it involves consumers who are trying to buy things, can benefit from other shoppers who share their specific needs, and find value in NLP-powered review and product information.
IIt’s a grand vision, to be sure. The task of the moment — getting it right in the beauty industry — is a large enough challenge for now. It’s unclear where exactly Mira stands; users are supposed to use the Mira app, but it’s available only for iOS at the moment. (You can also search directly from the AskMira site.) And although the company said it’s raised capital from multiple sources, including “Unilever Ventures, e.ventures, Founders Fund, 14W, Great Oaks Venture Capital, and more,” it’s curiously refused to share the amount. (Update, 10/15/19, 8:11pm PT: Mira has now shared that it raised $4.53 million.)
Author: Seth Colaner
Source: Venturebeat
