HN Showcase
Back to browse
GitHub Repository

Score your feed with EmbeddingGemma — Chrome extension + fine-tuning pipeline

6 starsTypeScript

Sift – Score your feed locally with EmbeddingGemma

by shreyask·Feb 24, 2026·1 point·0 comments
Visit ProjectView on HN

AI Analysis

●●SolidNiche GemBig Brain

Local feed scoring with fine-tuning, but Nitter/filter bots/feed algorithms preexist.

Strengths
  • •Runs entirely client-side via WebGPU + WASM; no data leaves browser.
  • •Taste profile + contrastive learning from thumbs up/down is clever signal extraction.
  • •25 built-in categories + CSV export pipeline enables real user-trained models.
Weaknesses
  • •Feed filtering is solved space (Reddit Enhancement Suite, feed algorithms, Nitter).
  • •Only 3 sites supported; utility depends on adoption across platforms author controls.
Category
AI/ML
Target Audience

Power users filtering noisy feeds; HN/Reddit/X users wanting local, privacy-first curation

Similar To

Reddit Enhancement Suite · Nitter · Goodlinks

Post Description

Sift through the noise. Score your feed with EmbeddingGemma, right in the browser.

Sift is a Chrome MV3 extension that runs fully client-side using WebGPU + Transformers.js. It scores posts on Hacker News, Reddit, and X and dims low-relevance items so the good stuff stands out.

Users can label items (/) to collect training data, export it as CSV, and fine-tune the model using the included Python pipeline.

Demo video: https://github.com/user-attachments/assets/83430c6b-b7fa-41c...

Repo: https://github.com/shreyaskarnik/Sift

Would love feedback!

Similar Projects