DEV Community

Cover image for 🦥
Sofia
Sofia

Posted on

🦥

#devchallenge #hf26challenge

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

Every morning, developers perform the same exhausting ritual: we open 12 browser tabs—HackerNews, Hugging Face blog posts, PyPI security advisories, and GitHub release notes—and skim through ~4,500 words of repetitive noise just to find the two sentences that actually matter before our coffee gets cold.

In the spirit of "Slow and Steady" (rewarding maximum developer laziness), I built SlothBrief 🦥.

SlothBrief is an autonomous, open-source morning intelligence studio powered by Google's open-weight Gemma model family (gemma3:4b (open-weight) via Ollama / gemma-3-27b-it). Instead of reading thousands of words across a dozen tabs, SlothBrief ingests noisy multi-source feeds, strips away 96%–99% of the PR fluff, calculates your "Skip Reading Score" (exact minutes of morning sleep gained), and speaks a crisp 30-second audio brief aloud—or tells you to go right back to sleep if nothing requires action.

✨ Key Lazy Features

  • High-Density Bento Studio UI: A zero-scroll, 2-column Bento grid inspired by Linear and Spotify that auto-hydrates on first load—showing the Morning Control Deck on the left and the Live Distillation & Audio Studio on the right.
  • Visual Browser Tab Crusher: Shows noisy source tabs (hugging_face_blog, pytorch_foundation, the_pragmatic_engineer, github_trending) with word counts visually crossed out and compressed into crisp cards.
  • 3 Configurable Laziness Levels:
    • 🦥 Ultra Sloth (10s read): Distills thousands of words of technical chaos into a single poetic 5-7-5 Haiku rendered in editorial serif typography.
    • ☕ Coffee Sipper (30s read): Generates exactly 3 high-impact executive bullets with zero marketing fluff.
    • 😴 Deep Nap (0s if quiet): Filters strictly for breaking changes or critical CVEs. If nothing is broken, it outputs: "💤 No action required. Go back to sleep."
  • Hands-Free Audio Player & Web Audio Synth: Streams voice briefings via ElevenLabs MP3 with automatic fallback to browser speech synthesis and a built-in Web Audio API pentatonic lofi synth + 48-bar reactive waveform equalizer so audio playback works smoothly in any environment.
  • DevTools-Grade Sentry Agent Waterfall: Custom sentry-sdk instrumentation (op="gen_ai.agent", op="tool.serpapi", op="gen_ai.summarize", op="tool.elevenlabs") rendered as a color-coded horizontal timeline waterfall alongside a 1-click 3-Scenario Benchmark Suite.

Demo

Code

GitHub logo sofiianowak / slothbrief

The Lazy Developer's AI Digest powered by Open-Weight Gemma

How I Built It

SlothBrief is built as a fast, zero-friction Python FastAPI single-service application serving an interactive Bento Studio frontend (TailwindCSS, Plus Jakarta Sans / Instrument Serif / JetBrains Mono, and HTML5 Canvas + Web Audio API).

System Architecture 

+-------------------------------------------------------------------------+
|                    SlothBrief 12-Col Bento Studio UI                    |
|  [Morning Control Deck]  [Tab Crusher]  [48-Bar Equalizer & Waterfall]  |
+------------------------------------+------------------------------------+
                                     | POST /api/brief & /api/benchmark
                                     v
+-------------------------------------------------------------------------+
|               FastAPI Orchestrator + Sentry Agent Tracing               |
|                  Root Span: op="gen_ai.agent"                           |
+---------+--------------------------+--------------------------+---------+
          |                          |                          |
          v                          v                          v
+-------------------+    +-----------------------+    +-------------------+
|   Ingestion Tool  |    |   Open-Weight Core    |    |  Audio Synthesis  |
| op="tool.serpapi" |    | op="gen_ai.summarize" |    |op="tool.elevenlabs|
+-------------------+    +-----------------------+    +-------------------+
| • Live SerpApi    |    | 1. Ollama (gemma3:4b) |    | • ElevenLabs MP3  |
|   Google News     |--->| 2. Gemma-3-27b-it API |--->| • Web Speech API  |
| • 12-Tab Multi-   |    | 3. Deterministic      |    | • Web Audio Lofi  |
|   Source Fallback |    |    Offline Distiller  |    |   Cadence Engine  |
+-------------------+    +-----------------------+    +-------------------+
Enter fullscreen mode Exit fullscreen mode

1. 3-Tier Open-Weight Gemma Engine (gemma_engine.py)

At the core of SlothBrief is Google's open-weight Gemma model family. To guarantee reliability across local developer machines, CI/CD test runners, and cloud deployments, I built a 3-tier cascading engine:

  1. Local Ollama (gemma3:4b / gemma2:2b): Runs 100% locally on http://localhost:11434 at $0.00 inference cost with complete data privacy.
  2. Cloud Open-Weight (gemma-3-27b-it): Uses the Google AI Studio API key to query open-weight Gemma models when deployed to serverless cloud containers.
  3. Deterministic Offline Distillation: Provides sub-20ms local signal extraction when running in air-gapped environments without a GPU.

2. Custom Sentry Agent Tracing (sentry_telemetry.py)

When an autonomous agent filters out 98% of your morning reading while you stay in bed, observability is essential. Every pipeline run records structured spans (gen_ai.agent, tool.serpapi, gen_ai.summarize, tool.elevenlabs) along with token counts, compression percentages, and minutes saved, exposing them via GET /api/traces to power the live UI waterfall chart.

Why Does Open Innovation Matter?

For a background "lazy" agent that summarizes thousands of words every single morning, open-weight models like Google's Gemma are essential over proprietary black-box APIs:

  1. Zero Marginal Cost for Daily Background Automation ($0.00): Scraping and summarizing 4,500+ words across a dozen feeds every morning quickly burns through millions of tokens. Running gemma3:4b (open-weight) locally means your morning laziness costs $0.00 forever with zero API billing anxiety.
  2. 100% Privacy for Internal Engineering Feeds: Developers don't just summarize public HackerNews threads—we also want morning digests of internal Jira tickets, private GitHub PRs, and security alerts. Open-weight local inference ensures sensitive code and logs never leave localhost.
  3. Portable & Verifiable Weights: Whether running locally via Ollama or deployed via render.yaml, open weights ensure consistent behavior without sudden proprietary model deprecations.

The Lazy Payoff

To measure the exact ROI of developer laziness (calculated at a technical reading speed of 250 words per minute), I included an automated CLI & UI benchmark suite (run_demo.py) across three real-world developer workflows:

Scenario Laziness Level Input Words (Tabs) Output Words Compression % Time Saved Latency AI Engine
Open Source AI & Model Releases Ultra Sloth (Haiku) 1,541 (4 tabs) 13 99.2% 6.1 mins 17 ms gemma3:4b (open-weight)
HackerNews Tech & Architecture Coffee Sipper (3 Bullets) 1,544 (4 tabs) 64 95.9% 5.9 mins 10 ms gemma3:4b (open-weight)
Python Security & PyPI Advisories Deep Nap (Action Items) 3,050 (8 tabs) 27 99.1% 12.1 mins 8 ms gemma3:4b (open-weight)

📈 Cumulative Lazy ROI:

  • Average Noise Eliminated: 98.1% (6,135 words across 16 tabs distilled into 104 high-signal words).
  • Morning Sleep Gained: 24.1 minutes saved per run (~146 hours saved per year).
  • Local Open-Weight Inference Cost: $0.00.

Prize Categories

  • Best Use of Gemma: Uses open-weight Gemma (gemma3:4b (open-weight) / gemma-3-27b-it) as the core distillation engine powering all three laziness modes (Haiku, 3-Bullet Coffee Sipper, and Deep Nap Action Verdict).
  • Best Use of Sentry Agent Tracing: Implements custom sentry-sdk spans (gen_ai.agent, tool.serpapi, gen_ai.summarize, tool.elevenlabs) visualized in a real-time DevTools-style waterfall chart inside the Bento UI.
  • Best Use of SerpApi: Powers multi-topic news discovery and live feed ingestion across developer topics.
  • Best Use of ElevenLabs: Turns Gemma's distilled brief into a hands-free morning audio stream paired with a 48-bar canvas waveform visualizer.
  • Best Use of Render: Includes a turnkey render.yaml Blueprint for 1-click cloud deployment.

Top comments (0)