AI Watch

Live site: https://Yatimai.github.io/ai-watch/

Autonomous agent that produces a daily AI briefing from 3 sources: HuggingFace Papers, GitHub Trending, and Simon Willison.

Runs every morning via GitHub Actions. Fetches, filters, enriches, summarizes. Not a one-shot notebook : a system that produces a fresh result every day.

How It Works

HuggingFace Papers API ---+
GitHub Trending (scrape) --+-- Top 3 per source -- LLM Agent + Tools -- Markdown Briefing
Simon Willison RSS --------+

The agent decides for each item whether it needs more context:

• Paper mentions a model -> searches HF Hub (downloads, license)
• Vague GitHub description -> reads the README
• Simon post with just a title -> fetches the full content

Quick Start

git clone https://github.com/Yatimai/ai-watch.git
cd ai-watch
python -m venv venv && source venv/bin/activate
pip install -e ".[dev]"
cp .env.example .env
# Add ANTHROPIC_API_KEY to .env
python -m src.cli run

Sources

1. HuggingFace Daily Papers: GET https://huggingface.co/api/daily_papers -> sort by upvotes -> top 3
2. GitHub Trending: scraping https://github.com/trending?since=daily (all languages) -> AI filter (LLM) -> sort by stars/day -> top 3
3. Simon Willison RSS: https://simonwillison.net/atom/everything/ -> top 3 recent posts

LLM

Single model: Claude Sonnet via Anthropic API. Two calls:

1. GitHub filter (JSON in -> JSON out, true/false per repo)
2. Enrichment + briefing (9 items -> optional tools -> markdown)

Prompts are in /prompts/.

Agent Tools

The enrichment agent has 3 tools:

• fetch_url(url) : read a web page (httpx)
• search_hf_models(query) : search a model on HuggingFace Hub (API)
• get_github_repo(owner, repo) : read README, stars, activity (GitHub API)

The agent decides which tools to call based on each item's content.

Briefings

Daily briefings are in briefings/. Each briefing has a log file showing the agent's tool-use decisions.

Project Structure

src/
  agent/
    graph.py        LangGraph StateGraph
    nodes.py        Node functions
    state.py        TypedDict state
    tools.py        Agent tools (fetch_url, search_hf, get_github)
  sources/
    huggingface.py  Fetch HF Daily Papers API
    github.py       Scraping GitHub Trending
    simon.py        RSS Simon Willison
  models/
    schemas.py      Pydantic models (Item, Briefing)
  utils/
    logger.py       Structured logging of agent decisions
  config.py         Settings (env vars + yaml)
  cli.py            CLI entry point
prompts/
  prompt-briefing.md       Enrichment + briefing prompt
  prompt-filter-github.md  GitHub AI filter prompt
briefings/                 Output: markdowns + logs (git tracked)
tests/

Stack

• Python 3.12
• LangGraph (orchestration)
• langchain-anthropic (LLM)
• httpx (HTTP)
• beautifulsoup4 (scraping GitHub Trending)
• feedparser (RSS Simon Willison)
• ruff (lint + format)
• mypy (types)
• pytest (tests, 86% coverage)

Dev

ruff check src/ tests/
mypy src/
pytest tests/ -v --cov=src

License

MIT