3 Prototypes: A Design and Engineering Exploration to Control Doom Scrolling (medium.com)

A design-and-engineering exploration called “3 Prototypes” proposes three pragmatic ways to curb doomscrolling while preserving creator value: Progressive Scroll Indicators (visual “cigarette bars” that fill with scroll and fade during pauses), Bottom Toasts (contextual, non-shaming prompts triggered at content boundaries using sentiment, time-of-day and session signals), and an AI Scroll Assistant (a lightweight companion that recommends pausing or switching content based on engagement patterns). The project includes iOS and web prototype code and argues for gentle pause moments to restore agency and promote deeper engagement rather than blunt time-based gating. Technically the work emphasizes low-latency, cost-conscious ML and simple client-side UX. Implementations use React + TypeScript + Framer Motion on web, SwiftUI on iOS, local caching for persistence, and SSE/WebSocket for live suggestions. Crucially, backend inference targets ONNX-powered models with <100 ms response times (optionally using Kafka for streaming), isolated suggestion microservices, and consented behavioral telemetry. Success metrics prioritize trust (pause rate, opt-in, accessibility), not raw minutes, and mitigations address under-signaling, annoyance, and perceived anti-growth. For the AI/ML community this is a useful case study in deploying small, real-time models for humane UX, balancing privacy, performance and platform adoption — with system-level OS integration flagged as the longer-term lever for widespread impact.