Dataset Discovery

Describe the data you need in plain language and get back a ranked shortlist of real Hugging Face datasets — each with a direct link and the facts that decide whether you can use it.

What it does

• Runs several focused Hub searches, so you see options one query misses.
• Opens each candidate and confirms modality, size, splits, and languages on its repository page — never the search snippet.
• Flags the licensing landmines up front: non-commercial, research-only, gated.
• Ranks by fit, breaking ties toward better docs and active maintenance.

What to send

Any of: task, domain, modality, language(s), size, license needs, exclusions. Partial briefs are fine — it states its assumptions and proceeds.

"Multilingual support conversations for intent classification — 50k+ rows, English plus two of German/French/Spanish, commercially usable."

What you get back

Markdown: the criteria searched, 3–8 ranked datasets, then notes on caveats and near-misses.

Searched: intent classification · en+de/fr/es · ≥50k rows · commercial-friendly

1. **[owner/support-intents-multi](https://huggingface.co/datasets/owner/support-intents-multi)**
   Labeled intents in all four languages, splits already defined.
   text · 180k rows · Apache-2.0 · updated 2 months ago

## Notes
- owner/other-corpus excluded: CC-BY-NC.

Good for

• Sourcing training or fine-tuning data for a specific task
• Finding an evaluation benchmark for a domain or language
• Clearing a dataset's license before it reaches production

Limits

Hugging Face Hub only, every fact from a Hub tool call. Licenses are a starting point for your legal check, not a substitute. It doesn't download the data, so it can't judge label quality, and anything a dataset card omits comes back as "not specified". If searches run and nothing fits, it says which criterion to relax; if the Hub is unreachable, it says so rather than reporting a false "no matches".