A personal daily digest of LLM/SLM inference-engineering news — vLLM, SGLang, TensorRT-LLM, llm-d, benchmarking, and related research — served as a three-tab, InShorts-style web page: Learn, Trending, LinkedIn Ideas.
Built to eliminate cross-platform scrolling for inference-engineering news and to surface both fresh developments and evergreen explainers worth writing about.
- A GitHub Actions cron job runs daily, fetching from arXiv, GitHub releases/PRs, vendor engineering blogs, and Hacker News.
- Items are filtered by keyword relevance (
config/keywords.yaml), then summarized and tagged into one of three tabs by an LLM call. - Results are written to
data/digests/and committed automatically. - The static web page in
web/reads the latest digest and renders interactive card feeds.
infer-daily/
├── README.md
├── .github/
│ └── workflows/
│ └── daily-digest.yml # daily cron workflow
├── config/
│ ├── keywords.yaml # editable keyword list
│ └── sources.yaml # source URLs and endpoints
├── pipeline/
│ ├── __init__.py
│ ├── fetchers/
│ │ ├── __init__.py
│ │ ├── arxiv.py # arXiv API fetcher
│ │ ├── github_releases.py # GitHub REST API fetcher
│ │ ├── rss_blogs.py # RSS engineering blogs fetcher
│ │ └── hackernews.py # HN Algolia API fetcher
│ ├── filter.py # keyword relevance filter
│ ├── summarize.py # LLM summarizer & tagger
│ ├── schema.py # Card & RawItem data schemas
│ └── run_digest.py # pipeline orchestrator
├── data/
│ └── digests/
│ ├── latest.json # copy of latest digest
│ └── YYYY-MM-DD.json # daily digest files
├── web/
│ ├── index.html # static three-tab app
│ ├── styles.css # dark-mode glassmorphism styling
│ └── app.js # renders digest JSON cards
└── tests/
├── test_filter.py
└── test_schema.py
uv pip install -r requirements.txt
# or: pip install -r requirements.txtuv run pytestpython -m pipeline.run_digestSupported LLM provider keys in .env or system environment:
NVIDIA_NIM_API_KEYOPENAI_API_KEYANTHROPIC_API_KEY
If no key is present, the pipeline gracefully falls back to deterministic heuristic card generation for local testing.
config/keywords.yaml— edit to add/remove topics tracked.config/sources.yaml— edit to add/remove RSS feeds and GitHub repos.
- Pipeline: Fully built with arXiv, GitHub, RSS blogs, and Hacker News fetchers.
- LLM Summarizer: Built with multi-provider support (NVIDIA NIM / OpenAI / Anthropic) and soft fallback.
- Frontend: Fully built static three-tab InShorts UI in
web/. - Automation: GitHub Actions daily cron workflow configured in
.github/workflows/daily-digest.yml.