Bouncer

Product Introduction

1. Definition: Bouncer is an AI-powered social media content moderation tool, categorized technically as a Semantic Feed Filter. It is currently available as a Google Chrome extension and a native iOS application, designed to interface directly with social media platforms—beginning with Twitter/X—to programmatically curate user feeds.

2. Core Value Proposition: Bouncer exists to shift the power dynamic between social media algorithms and individual users. By utilizing a Small Language Model (SLM) to perform real-time sentiment and context analysis, Bouncer allows users to reclaim their attention from engagement-optimized algorithms. It enables the filtering of "toxic" content patterns, specific ideological topics, or undesirable linguistic tones, effectively "healing" the user’s feed through modified engagement signals.

Main Features

1. Context-Aware Semantic Filtering: Unlike traditional keyword-blocking tools that rely on exact string matching, Bouncer utilizes the Qwen 3.5 4B open-source Large Language Model (LLM) to understand intent and nuance. This technology allows users to filter content based on abstract concepts such as "AI doomerism," "crypto," or even subjective emotional tones like "arrogance" and "pessimism." The model classifies tweets in real-time, ensuring that content which matches the semantic profile of a user's filter is hidden before it can impact the user's attention.

2. Algorithmic Feedback Loop Modification: Bouncer functions as an intermediary layer between the user and the platform's recommendation engine. Because social media algorithms (like Twitter's "For You" tab) optimize based on what a user dwells on or interacts with, Bouncer’s act of "bouncing" (hiding) posts prevents the native algorithm from recording engagement on those topics. Over a period of days, this lack of engagement trains the platform's primary algorithm to stop suggesting similar content, resulting in a permanent, native improvement of the feed.

3. High-Performance Edge Computing Roadmap: Currently, Bouncer processes classification tasks via Imbue’s dedicated data centers to maintain the speed required for seamless scrolling. However, the architecture is designed for future migration to local execution. By leveraging the dormant GPU power in modern laptops and iPhones, Bouncer aims to move the inference process entirely on-device. This technical direction ensures low-latency performance while moving toward a "personal computing" model where AI-driven filtering remains entirely under the user's local control.

Problems Solved

1. Pain Point: Algorithmic Hostility and Doomscrolling: Traditional social media algorithms are optimized for "attention," which often translates to promoting content that triggers anger, anxiety, or addiction. Users frequently feel trapped in a "cesspool" of negative content that they cannot manually escape. Bouncer provides a technical "firewall" against these psychological triggers.

2. Target Audience:

• Professional Knowledge Workers: Individuals who use Twitter/X for real-time information but need to filter out the high noise-to-signal ratio of professional "arrogance" or repetitive industry hype.
• Political Enthusiasts and Researchers: Users who need local political updates but want to avoid the vitriol and inflammatory tone often associated with election cycles.
• Mental Health-Conscious Users: Individuals seeking to reduce "doomscrolling" and the negative psychological impacts of platform-driven outrage.
• Privacy-Conscious Tech Early Adopters: Users who prefer open-source model implementations (Qwen) and personal control over big-tech data center processing.

3. Use Cases:

• Industry Noise Reduction: A developer filtering out all "crypto" or "NFT" discourse without losing access to general tech news.
• Tone Regulation: A user hiding all posts classified as "pessimistic" or "defeatist" during high-stress news cycles.
• Niche Topic Exclusion: Automatically removing specific discourse, such as "AI doomerism," while keeping broader "Artificial Intelligence" updates.

Unique Advantages

1. Differentiation: Traditional moderation tools are "dumb" and easily bypassed by slang or slight spelling variations. Bouncer’s LLM-based approach understands the meaning of a post, making it impossible for "garbage" content to bypass filters simply by changing keywords. Furthermore, unlike "ad-blockers" which only remove static elements, Bouncer actively reshapes the underlying recommendation engine of the platform it inhabits.

2. Key Innovation: The specific use of a 4-billion parameter model (Qwen 3.5 4B) strikes a critical balance between "intelligence" (the ability to recognize complex human emotions like arrogance) and "latency" (the ability to process a feed at the speed of a human thumb scrolling). This represents a shift toward "Small AI" that is specialized for user-side empowerment rather than corporate-side data mining.

Frequently Asked Questions (FAQ)

1. How does Bouncer differ from Twitter’s native "Muted Words" feature? Twitter’s native mute feature only hides specific words or hashtags. Bouncer uses a Large Language Model to understand the context. For example, if you mute the word "Apple," Twitter hides everything about the fruit and the company. Bouncer can be told to hide "arrogance," and it will analyze the tone of the tweet to determine if it should be hidden, regardless of the specific words used.

2. Can Bouncer actually change what the Twitter algorithm shows me? Yes. Twitter’s algorithm learns from your behavior. If you don’t see a post (because Bouncer hid it), you won’t click it, like it, or dwell on it. Twitter interprets this lack of engagement as a lack of interest, eventually leading the native algorithm to stop serving you that type of content altogether.

3. Is Bouncer available on mobile devices and other browsers? Bouncer is currently available as a Chrome extension and an iPhone app. The developers at Imbue have announced plans to expand support to Safari, Firefox, and other social feeds including Reddit and LinkedIn in the near future.

4. Does Bouncer process my data in the cloud? Currently, classification is handled in Imbue’s data centers using the Qwen 3.5 4B model. However, the technical roadmap includes moving this processing to the local GPUs of your laptop or phone, ensuring that your feed filtering remains private and under your direct control.