Open Data Finder
In an 18-need benchmark every top link resolved, against 81% for a frontier LLM with web search, at about 1/90th of the cost. A small, fast LLM plans the searches and eight open catalogs supply the datasets.

Highlights
- Takes a need in plain words, like daily rainfall by district in Gujarat, and returns real open datasets.
- A small, fast LLM plans the searches; eight open catalogs answer, including Hugging Face, Zenodo, NASA Earthdata and Harvard Dataverse.
- Every link is fetched before it is listed, and code blocks anything a catalog did not return.
- 18-need benchmark: every top link resolved, versus 81% for a frontier LLM with web search, at about 1/90th the cost.
- Lives on tensortapri.org as a free community tool, on Cloudflare Pages Functions with a per-visitor limit.
The full story
Finds real open datasets for a need in plain words, like daily rainfall by district in Gujarat, as a free tool on Tensor Tapri. A small, fast LLM plans the searches, eight open catalogs answer (Hugging Face, Zenodo, NASA Earthdata, Harvard Dataverse and more), every link is fetched before it is listed, and code stops the model adding anything a catalog did not return. In an 18-need benchmark every one of its top links resolved, against 81% for a frontier LLM with web search, at about 1/90th of the cost.