AbhavyaManchanda

Posted on Oct 25

SlothDigest: The Zero-Effort Local RSS & Newsletter

#devchallenge #hf26challenge #ai #python

Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady Submission 🦥

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady

🦥 What I Built

In 2026, the developer ecosystem suffers from an acute attention crisis: infinite dopamine feeds, aggressive product marketing disguised as engineering blogs, and firehoses of unread newsletters.

Most of us subscribe to 20+ RSS feeds and developer newsletters with great intentions, only to face "inbox bankruptcy" and doomscroll through promotional fluff.

SlothDigest was built to restore digital sanctuary. Inspired by the calm, deliberate philosophy of the sloth—"Slow and steady wins the signal race"—SlothDigest is a 100% local-first, zero-cloud RSS and newsletter curator.

Instead of demanding hours of reading:

Unified Local Ingestion: Ingests your RSS/Atom feeds (Hacker News, GitHub Engineering, Simon Willison) and local .eml email newsletter folders.
HTML Boilerplate Stripping: Purges tracking pixels, sponsor headers, and navigation chrome.
The Sloth Filter Engine (Local Ollama): Passes clean prose to an open-weight local LLM (llama3.1:latest, llama3.2:1b, or qwen2.5:3b) running privately on localhost:11434.
Ruthless Fluff Elimination: Discards 85%+ marketing noise, promotional pitches, and SEO listicles.
Rejection Transparency Audit: Logs every dropped article and the exact AI reasoning behind the drop.
1-Minute Spoken Audio Briefing: Synthesizes a daily editorial audio briefing MP3 with an animated 28-bar waveform player.

📸 Demo & Interface Walkthrough

SlothDigest is designed with an editorial, minimalist aesthetic inspired by Linear and Readwise Reader (deep slate #06090e, charcoal, muted forest green, and amber accents—no generic neon AI gradients).

Key Views:

Daily High-Signal Digest: Top 3–5 high-value takeaways featuring a high-impact TL;DR, core technical mechanics, impact statement, and reading time.
Interactive Audio Briefing: Custom player with live animated waveform bars, speed toggle (1.0x – 2.0x), and scrub controls.
Unified Intake Dashboard: Manage RSS URLs and point to local ./newsletters email folders with real-time sync indicators.
Rejection Log View: An audit trail showing what was dropped (e.g. “Promotional launch pitch”, “Low information density”).
Settings & Prompt Studio: Fine-tune Ollama models, filtering strictness, and morning schedule times (e.g. 7:00 AM).

🏗️ Architecture & Tech Stack

SlothDigest runs entirely on your local machine with zero telemetry or subscription lock-in.

flowchart TD
    subgraph Intake["1. Local Intake"]
        RSS["RSS / Atom Feeds\n(Hacker News, GitHub Blog)"]
        EML["Local Newsletters\n(./newsletters/*.eml)"]
        Cleaner["HTML Boilerplate Stripper\n(Decomposes scripts, navs, footers, tracking pixels)"]
        RSS --> Cleaner
        EML --> Cleaner
    end

    subgraph Core["2. The Sloth Filter Engine"]
        Cleaner --> Prompt["Custom Prompt Studio\n('Drop marketing fluff, keep engineering breakthroughs')"]
        Prompt --> Ollama["Local Ollama API\n(llama3.1 / llama3.2 on localhost:11434)"]
        Ollama --> Decision{"Signal Score >= 6.0?"}
        Decision -- "Reject (< 6.0)" --> RejectionLog["Rejection Audit Log\n(Reason, Dropped Score, Analysis)"]
        Decision -- "Keep (>= 6.0)" --> Curator["High-Signal Curator\n(Top 3-5 Takeaways: TL;DR + Impact)"]
    end

    subgraph Audio["3. Audio Synthesis"]
        Curator --> ScriptGen["Briefing Script Generator"]
        ScriptGen --> TTS["Neural TTS (edge-tts / pyttsx3)\nGenerates briefing_digest.mp3"]
    end

    subgraph UI["4. Digital Sanctuary Interface"]
        TTS --> AudioPlayer["Animated Waveform Player"]
        Curator --> ReadingView["Distraction-Free Reading Sanctuary"]
        RejectionLog --> AuditView["Filter Transparency Dashboard"]
    end

Technology Breakdown:

Backend: Python 3.10+, FastAPI, SQLite (slothdigest.db), Uvicorn.
Local AI: Ollama running open-weight models (llama3.1:latest, llama3.2:1b, qwen2.5:3b) over local REST API with a heuristic fallback engine.
Feed & Mail Ingestion: feedparser, beautifulsoup4, Python standard library email (RFC 822 / .eml).
Audio Engine: edge-tts (neural voice synthesis) with offline pyttsx3 fallback.
Frontend: React 18, Vite, Tailwind CSS, Lucide icons.

🏃 Quickstart: Running SlothDigest

1. Clone & Setup

git clone https://github.com/AbhavyaManchanda/SlothDigest.git
cd SlothDigest

2. Install Dependencies

# Python backend
pip install -r requirements.txt

# React frontend
cd frontend
npm install
npm run build
cd ..

3. Ensure Local Ollama is Running

ollama pull llama3.1
# or a lightweight 1B model:
ollama pull llama3.2:1b

4. Launch the App

python run.py

Open http://localhost:8000 in your browser.

🐢 Why "Slow and Steady" Matters

The theme for Hacktoberfest 2026 Week 3 is "Slow and Steady", and it speaks directly to why SlothDigest was created:

Combating Attention Fragmentation: High velocity is not high signal. By intentionally bottlenecking content intake to a daily 3–5 item digest, we trade superficial skimming for deep architectural understanding.
Open-Weight AI Belongs to Everyone: You don't need a cloud subscription to have high-tier AI curation. A local 1B or 3B model running on consumer hardware can distill news without harvesting your browsing habits.
Auditability Over Black Boxes: Most AI feeds hide what they censor. SlothDigest provides an explicit Rejection Log so the reader remains in full control of their information diet.

🔗 Repository & Source Code

GitHub Repository: https://github.com/AbhavyaManchanda/SlothDigest
License: MIT License

🤖 AI Disclosure

SlothDigest was designed and developed with the assistance of agentic AI coding pairing (Gemini 3.8 Flash). The core application integrates local open-weight models (llama3.1) via Ollama for local-first inference.

DEV Community