01 / Safety by Design
Community Sift combines advanced machine learning and next-generation AI to allow communities of every size to identify, classify, and moderate user generated content.
Sift’s AI-enabled capabilities escalate the toughest cases to human moderators, significantly increase accuracy in assessing nuanced communication, and provide data-driven recommendations.
100B+
Human interactions classified, filtered, and escalated each month
88%
Reduction of moderator workload with AI-enhanced decisioning
96%
Reduction in reported toxic incidents via built-in features
02 / SIFT B.D. (Before design)
For years, Community Sift existed as a web application without a proper front-end. Because it had no cohesive user experience or interface—not even a navigation—onboarding required a white glove experience and a customer success team available at all hours to hold customers’ hands as they navigated a bespoke and esoteric experience.
I was brought in as the UX/UI project was underway and the design language had already begun to be established. My job was to manage the design of the product and help build out Figma styles, components, and prototypes.
Content Queues are set up to organize filtered UGC for moderation, based on the particulars of your community.
Once a UI had been built for the existing functionality, we began the work of designing Next Generation Moderation. NGM would require us to reassess every aspect of the product, including look and feel, which meant building out many new components, as well as a dark theme.
03 / Sift Design System
As we pivoted away from the old platform of spaghetti code and into NGM, our most senior frontend engineer and I persuaded the head of product to invest time into building out a design system. This would prove crucial to enabling us to build a more cohesive experience at scale — a need that became urgent once our team was acquired by Microsoft in 2021.
I owned the design system and tapped our best front end engineer to help me build and maintain our growing library of components and styles using Figma, Chromatic, and Storybook.
Having initially built a design system within Figma, we found it to be a struggle to get engineers to use it. Our engineering team either didn’t know where to look for specs or just didn’t feel comfortable in Figma and avoided it. Therefore, they continued to build as they always had—each dev with their own process. When we finally established the same design tokens and complex components within Storybook, we were able to offer complete prototypes complete with design tokens, presented in a way that was comfortable and that made sense to them. As a result, the consistency and usability of our interface improved exponentially.
All existing design tokens, components, complex patterns, and documentation were re-established within Storybook.
04 / Next Generation Moderation: TEXT & IMAGES
While the original platform relied upon a combination of AI and human input, the need to scale quickly resulted in a decision to lean more heavily upon AI. In doing so, the most difficult decisions would still be escalated to human moderators for review while the vast majority of decisioning—the obvious allow or reject decisions—would be handled by AI. Thus lessening the workload for human mods while also protecting them emotionally and cognitively. Enter: Next Generation Moderation (NGM). Users would now be able to moderate text and images using a highly customizable interface.
Custom Policy Guides
Set filters to align with your community guidelines (e.g. no hate speech, violence, level of tolerance for vulgarity) and reinforce your Terms of Use. Users can choose from their chosen presets or get even more granular using sliders.
AI Moderation
Enable AI to moderate content in a queue based on the platform’s policy settings. All AI-moderated content is automatically sent to a queue for review by a human moderator.
NGM Images
Quickly scan through and decision images queues using a single click or hot keys.
Curated topics and custom risk sliders allow you to prescribe an ideal user experience while optional image blurring protects moderators’ wellbeing.
Filter queues by date topic, date, moderator, guest, category, or subcategory.
Need a closer look? Users can opt to view images one at a time, viewing any associated detail.
When selecting imagery for empty states, rather than go the typical route of corporate illustration, I chose adorable images of animals combined with playful copy. Not only a more joyful approach, but research shows that such images trigger the brain’s reward centers. “Cute features not only make objects more user friendly and approachable, but also induce careful behavioral tendencies in the users…” [1] “…researchers suggest that cute images may be helpful in improving performance for jobs that demand significant attention to detail, like air traffic control or software programmers.” [2]
Users are able to create “Escalate All” queues, which automatically redirects all content to AI for moderation.