Good work.
Meet the right support.

AI downloads public model files from this site, then compares meaning locally. Best for English descriptions. Keyword matching works without the model.

Thematic fit and readiness are separate. The headline fit grade measures supported alignment between your description/tags and a program’s documented focus. 80–100 means Strong thematic fit; 60–79 Relevant themes; 40–59 Partial overlap; below 40 Limited overlap. A disclosed rubric combines specific shared-term evidence and local semantic similarity. It is not a probability of funding or a measured accuracy percentage. Open the fund details for the exact rubric.

Next-step badges show eligibility conflicts, questions to confirm, funding-type compatibility, application routes and timing. Missing applicant details or a closed cycle affect shortlisting, rather than lowering thematic fit. The default order prioritizes the next practical step, so it is not a simple descending sort by thematic fit. All 699 records remain available.

These are research leads, not award probabilities or eligibility approvals. Every record has a research disposition, with program terms separated from source-only overviews and unresolved checks. Sourced executable checks cover a subset of programs; other documented requirements remain manual. Geography mentions are never treated as an exhaustive eligibility list. Leadership, budgets, local boundaries and other conditions need review. Inquiry routes have their own filter and do not count as open applications.

Model: Mixedbread embed xsmall v1 (Apache 2.0), quantized and pinned. Uses a browser worker and WebAssembly, following the Artizen approach. The browser may cache public model files; project embeddings are discarded when stopped. This page blocks outbound inference connections.