briefing-32

A 32B-class AI-news briefing the maker runs every 2 hours.

Build Small Hackathon entry (Backyard AI track). Down-ported from the production ai-news-agent cron (Groq Llama-3.3-70B → WhatsApp) onto Qwen3-32B served by Hugging Face Inference Providers.

Pipeline: RSS + HN + arXiv + GitHub → cheap relevance filter → graded 0–10 ranker → readable digest. Two open-weight model calls, no 70B cloud round-trip required.

Controls

Window (hours back)

Production runs every 2hr — match that for the authentic story.

1 72

Sources

rss hn arxiv github

Model (≤32B params)

Default Qwen3-32B. Swap to Qwen3-30B-A3B for faster MoE inference.

HF_TOKEN (optional — reads env if blank)

Run stats

no run yet

Digest

Click Run briefing to fetch the last N hours of AI news, rank it on a ≤32B model, and render a readable briefing.

Ranked items

Build Small Hackathon · Backyard AI track. Apache 2.0. Code: github.com/MukundaKatta/briefing-32