Speech Technology Magazine

MindMeld Launches Deep-Domain Conversational AI

MindMeld's Conversational AI platform simplifies onboarding and integration to other applications, like Siri and Cortana.
By Tye Pemberton - Posted Nov 18, 2016
Page1 of 1
Bookmark and Share

Artificial intelligence company MindMeld launched a speech-enabled Conversational AI platform this week to provide what it calls "deep-domain" AI, through the production and development of dynamic knowledge graphs in the vein of leading search engines.

Tim Tuttle, MindMeld's founder and CEO, says the product has been in production for the company's entire five-year lifespan. "We started MindMeld {in 2011] specifically with the intention of building an AI platform like this," he says. "We released an open-ended version about 12 months ago, but this is what we've been building toward from beginning."

MindMeld Conversational AI is unique in that the onboarding is provided by MindMeld through the deep-domain dynamic knowledge graphs on which the AI functions. These graphs, designed to provide the AI with client-specific information, act as both a roadmap for AI response and a starter pack of data. The knowledge graphs can include information such as product details, frequently asked questions, specialized vocabulary, and relational information.

Because of this client-tailored approach to the AI knowledge base, there is a discussion and training period during which MindMeld and clients optimize the Conversational AI for launch.

"There's essentially no such thing as out-of-the-box training data for our product," Tuttle says. "We help our customers create the initial set of data. We use a lot of crowdsourcing techniques, so the process is not immediate."

While the process might not offer "hello-world" at the push of a button, the onboarding period for such a robust conversational AI is still short. Tuttle suggests that onboarding only takes several weeks, with smaller onboardings lasting as little as a few days.

After the initial process to get clients to "hello-world," MindMeld's Conversational AI platform continues to develop and adapt based on user input and interaction. MindMeld treats collected data as the property of individual clients, so the data is corralled rather than aggregated centrally, effectively producing a unique product for each of MindMeld's clients while ensuring that their data and approaches remain proprietary.

While MindMeld uses home-grown speech recognition in the Conversational AI, the platform is already interoperable with most commercial intelligent assistants, like Apple's Siri, Amazon's Alexa, and Microsoft's Cortana. "We are able to leverage their own robust speech recognition, provided our clients' users are operating through those channels," Tuttle says. "What's especially exciting in the current ecosystem is that larger consumer companies have speech recognition as a given in their tech, and now we are developing interoperability between intelligent assistants. A user can ask Siri to engage with one of our clients, and then our AI will take over the conversation."

Tuttle also notes that monitoring goes hand-in-hand with the platform. "In many cases, MindMeld manages the infrastructure, and we're always at-the-ready to report problems, issues, or developments," he says. "While we also offer the option for our clients to host our platform locally, so far they have preferred to opt for us."

MindMeld already has the high-profile client Uniqlo using the Conversational AI as an assistant in Facebook Messenger. "They're forward-looking and seeking to engage customers across all conversational platforms, even those beyond traditional customer service channels," Tuttle says.


Page1 of 1