nvidia-developer-firehose - SKILL.md Agent Skill

name: nvidia-developer-firehose description: | Real-time AI-ECOSYSTEM firehose (v3 multi-source, May 2026). Polls 12 Atom/RSS feeds every 30 min: NVIDIA (developer-blog + main-blog + newsroom), hyperscalers (Azure, AWS, AWS-ML, Meta-Engineering), AI labs (OpenAI, DeepMind, Hugging Face), and neoclouds (CoreWeave, Together AI). For each new post, uses HEURISTIC EXTRACTION + yfinance.Search to auto-resolve every mentioned company → US ticker (no hand-maintained name→ticker dict), with a persistent ticker cache that learns over time. Surfaces names via Telegram tagged by source, separated into 🎯 portfolio-tracked tickers vs 🔍 newly discovered tickers. Why it matters: hyperscalers + NVIDIA + AI labs publicly name 800V HVDC, CPO, optical, power, and custom-silicon partners — the forward-looking design ecosystem that re-rates weeks later when sell-side picks it up. (AMD has no public RSS — covered separately by SEC 8-K strategic-partner-firehose.) Triggers in English ("nvidia developer firehose", "ai ecosystem firehose", "ai partner monitor", "hyperscaler blog scraper") or Chinese ("NVIDIA Developer 爬虫", "AI 生态 firehose", "超大厂博客实时", "英伟达 + 超大厂 + AI 实验室博客"). Do NOT trigger for: generic web scrapers, non-equity contexts, or individual blog reading.

NVIDIA Developer Firehose

Why this exists

NVIDIA's Technical Blog (developer.nvidia.com/blog) is where NVDA publicly names the partners it's designing the next generation around — 800V HVDC chip suppliers, CPO optical module makers, custom XPU/ASIC partners, networking/server OEMs, hyperscaler/neocloud customers. When a partner gets named here, it's typically weeks before the Substack / sell-side picks up on it. Forward-looking edge.

This firehose:

Polls the Atom feed every 30 min.
For each new post, runs smart heuristic extraction + yfinance.Search to auto-resolve every capitalized noun phrase → US-listed ticker.
Persistent cache (nvdev_ticker_cache.json) learns over time so newly-named partners get added automatically; never needs hand edits.
Sends a Telegram alert separating 🎯 portfolio-tracked mentions (high signal) from 🔍 newly discovered tickers (medium signal).

Architecture

ATOM FEED                  CANDIDATE EXTRACT             RESOLVE
─────────                  ─────────────────             ───────
developer.nvidia.com  →    capitalized 1-4 word     →    cache HIT  →  done
/blog/feed/                phrases, minus              cache MISS →  yf.Search()
                           STOPWORDS + ALIAS                          name-overlap
                           expansion                                  US exchange?
                                                                      → cache + emit

No hardcoded name_to_ticker dict. A small TRACKED_TICKERS set (your portfolio universe) only affects whether a hit shows up as 🎯 or 🔍. Resolution itself is fully automatic.

ALIAS map handles only ambiguous short forms (e.g. MPS → Monolithic Power Systems) so the yfinance search returns the right company.

STOPWORDS filter strips NVIDIA's own products, calendar/structural words, and common tech jargon (CUDA, TensorRT, Blackwell, GTC, etc.).

Negative cache: words that resolve to nothing (or non-US) are also cached, so we never re-look up junk.

Files

scripts/nvdev_firehose.py — the engine
(in the runtime repo)
- scripts/nvdev_state.json — seen-post ids (5k cap, dedup)
- scripts/nvdev_ticker_cache.json — name → {ticker,exchange,name,source}

Workflow

The runtime repo provides a GitHub Actions workflow that runs the engine on a */30 * * * * cron with TELEGRAM_BOT_TOKEN/TELEGRAM_CHAT_ID as secrets. First run silently seeds the state (no Telegram avalanche).

Tuning

Env var	Default	What it does
`MAX_ALERTS`	8	Cap Telegram alerts per run
`MAX_LOOKUPS`	40	Cap yfinance.Search calls per article (cache absorbs rest next run)
`TEST_MODE`	(unset)	If `1`, log to stderr; no Telegram, no state/cache write
`NVDEV_USER_AGENT`	(generic)	Override the polite UA string

Editing the cache

The cache JSON is human-editable. To override a wrong resolution or add an alias by hand:

{
  "some company": {"ticker": "ABCD", "exchange": "NMS", "name": "Some Company Inc.", "source": "manual"},
  "wrong match name": null
}

null = negative-cached, will be skipped on future runs.

What an alert looks like

🟢🟪 NVIDIA Developer — new post
NVIDIA Transitions to 800V HVDC for 1MW AI Racks
作者: Mathias Blake et al
发布: 2025-05-20

🎯 组合内标的(高信号):
  • Infineon → IFNNY
  • MPS → MPWR
  • Navitas → NVTS
  • STMicroelectronics → STM
  • Texas Instruments → TXN
  • Eaton → ETN
  • Schneider Electric → SBGSF
  • Vertiv → VRT
  • TSMC → TSM
  • Coherent → COHR
  • Lumentum → LITE
  • Marvell → MRVL

🔍 其他识别到的可交易标的:
  • ROHM → ROHCY  (ROHM Co., Ltd.)
  • Flex Power → FLXP  (Flex-Power Inc.)

https://developer.nvidia.com/blog/...

Related skills

strategic-partner-firehose — SEC 8-K / 13D partner detection (downstream)
pr-wire-firehose — vendor newsroom RSS (parallel signal source)
insider-firehose — Form 4 buys (corroborating signal)