Wit.ai is Facebook open natural language processing (NLP) platform. It’s the tool that turns what users say (voice or text) into action. This allows developers to enable voice interface for their chatbots, mobile app, smart home, and wearable devices – including Facebook Portal and Oculus. Wit.ai supports over 200,000 developers around the world with over 132+ languages.
To revitalized and modernize Wit.ai, I was brought in to lead the redesign from the ground up to compete with Amazon Alexa and Google Dialogflow.
Role & Responsibilities
Senior Product Designer, Facebook
Tooling vision, information architecture, interaction design, research
Voice and speech recognition market is expected to grow at 17.2% and worth $26.8 billion by 2025. End-users will expect to interact with products with voice - being apps, IoT, or wearables. So developers need to support the VUI experience to be competitive in the market.
From user research and interviews, developers see VUI as just another feature that adds value while their main job to be done is still building the best core experience of the product they own.
So instead of spending time learning about the theory behind it, they need tools that make VUI development easy, reliable, and adaptable to their use cases. Additionally, many developers do not want to be locked into a specific platform like Alexa or Google Assistant which has a lot of overhead.
After launched, majority of the users also said that the new interface was their favorite change and praised how easy it was to use. The developer engagement was extraordinary as well. Wit.ai had the highest participation in Facebook global hackathons - beating Spark AR and even Messenger.
increased in Community Engagement
increased in userbase
satisfactory rating
Assistant Developer platform as part of Wit.ai announcement at F8 2021
With Wit.ai, developers can create VUI in less than 20 mins by providing the few ways people would say to activate the command. Wit ML algorithms will figure out what are the intents, providing expansive slots, and export out a simple JSON. Developers then can query this through the API and enable the voice command free of charge.
Wit.ai is a great platform for any solution that involves NLP and was really user-friendly. It was a good experience exploring it.
— Wit.ai Jam participant
Mario level 1-1 is famous for the design that teaches people the tool to master in the very beginning through play. And my goal is to use the same principle with NLP.
The "Understanding" section of the platform starts by letting people put in a command - something that everyone knows how to do. Then ask what "intent" or what desired outcome of that command is. And that's all they need to start.
The UI also allows easy discoverability of more advanced concepts like slots or traits later. This allows developers to build the right mental model on how NLP objects fit together.
The hardest part of NLP is to predict what command people will use. To let developers improve this quickly, the UI shows the data stream of what actual users said during the sessions. Developers can see how well their NLP model fits the users' expectations and quickly iterate right on the UI.
The biggest learning moment was honestly how easy it was to integrate wit.ai - I had already created a basic language model via Voicify to use on the other supported platforms like Alexa and Dialogflow. Getting that working in wit was unbelievably easy.
— A Voice Powered Controller to Level the Playing Field
Having a great developer platform isn’t just having a good GUI but also great API, documentation, and robust community. In the design, I made sure this information is easily accessible throughout the tool with a just-in-time learning experience to reduce the cognitive load.
On that note, this video is sponsored by Wit.ai. I’m just kidding. That’s a lie. But it’s F*cking should be. Because Wit.ai is the Sh*t.
— MatthewHallberg, My Boss made me Code an AI Assistant...
