InboxVudu uses NLP to help you focus on the emails that matter

A text-analysis startup called Parakweet (whose initial product focused on book recommendations) has launched a new application, called InboxVudu, that’s designed to help users reduce the stress of email by showing them just the messages that need their attention. And while it turns out that no amount of curation can really help ease the email burden of a technology journalist today, the app might work very well for other folks.

InboxVudu works by analyzing the text of a message and figuring out if the sender is asking something from the recipient — “I need an answer to that big question,” or “Please RSVP by Feb. 13” or something along those lines. At the end of the day, users receive an email from InboxVudu showing them the message that need their attention. From that email, as well as an associated web application, users can reply to emails, mark them as “resolved,” flag false positives and even mute the sender.

In Gmail, at least, the messages also find their way to an InboxVudu-labeled folder users can peruse them at their leisure.

A sample screenshot of an InboxVudu "digest."
A sample screenshot of an InboxVudu “digest.”

Parakweet co-founder and CEO Ramesh Haridas explained in an interview that the app works with about 90 percent accuracy today, based on internal testing, and that he hopes it will get even smarter as the company adds more signals into its models. For starters, there’s all the interaction data that users will generate by replying to messages and flagging false positives, which Parakweet can use to train the system both individually and at an aggregate level. Haridas also suggested the application might someday prioritize emails from people to whom users respond very quickly, or consider a sender’s job title or other measures of “global importance.”

I suggested InboxVudu could be really valuable as a way of helping users understand their “email graphs,” if you will, and Haridas agreed. He said the company is considering offering users’ statistics about their activity and which of their email contacts are the most important. However, he made sure to add with regard to privacy, “It’s all being processed by machines, so it’s never seen by a human being.”

Those feature and algorithmic improvements are still a way out though, but I’ve been using the first iteration of InboxVudu for about a week. And although it works as advertised — I’ve noticed few if any false positives — seeing 20 nicely bundled PR pitches doesn’t make it any easier to read them all or reply to them all. I’d consider muting the senders, but the nature of PR is that sometimes pitches are compelling and sometimes they’re not, even from the same person.


One really nice thing about InboxVudu, though, are the “follow-up” messages it displays — those where I’ve asked something from someone else and am still awaiting a response. I’m prone to being disorganized, so any reminders of the various stories or projects I’m working on, especially ones involving other people, are helpful.

If there’s one thing I’m confident about, though, it’s that programs like InboxVudu will continue to get better as the field of artificial intelligence continues to improve. As the speakers at our upcoming Structure Data (March 18-19 in New York) and Structure Intelligence (Sept. 22-23 in San Francisco) conferences demonstrate, advances in AI are happening fast, especially in fields such as language understanding and personal assistant technologies.

If Parakweet’s Kiam Choo, who studied under deep learning guru Geoff Hinton at the University of Toronto, can figure out a way to make to make email substantially less burdensome even for people like me with 29,000 unread messages, then the more power to him.