Lookalike

When you have dozens of browser tabs open, manually organizing them is time-consuming and tedious. Traditional grouping methods rely on keywords or domains, but they miss the actual meaning behind the content.

The Problem

Tab overload is real. You're researching, and suddenly you have 40 tabs open. Some are about JavaScript frameworks, some are Python docs, others are random Stack Overflow questions. Manually organizing them? Nobody has time for that.

How It Works

Lookalike uses semantic AI to understand what each tab is actually about, not just keyword matching:

1. Content Extraction - Extracts meaningful text from each page: titles, meta descriptions, headings, and key phrases
2. Semantic Embeddings - Converts content into 384-dimensional vectors using the all-MiniLM-L6-v2 model
3. Similarity Matching - Compares tab embeddings using cosine similarity (threshold ≥ 0.45)
4. Intelligent Naming - Groups get meaningful names based on shared themes

For example, tabs about "JavaScript tutorials", "React documentation", and "Node.js guides" will be grouped together as "Web Development", even though they share few common keywords.

Privacy First

All AI processing runs locally in your browser. Your browsing data never leaves your computer. The model (~23MB) is downloaded once and cached for future use.

Technical Details

• Model: Xenova/all-MiniLM-L6-v2 via transformers.js
• Embedding Size: 384 dimensions
• Similarity Threshold: 0.45 (cosine similarity)
• Minimum Cluster Size: 2 tabs

Usage

• Analyze All: Process all open tabs and group them automatically
• Regroup: Re-cluster tabs after opening new ones
• Ungroup: Remove all groups

This project showcases how modern browser extensions can leverage on-device AI for intelligent features without compromising user privacy.