AI notice
Short version: no model decides anything you see. Models fill forms and extract fields from official text. Every field they produce is stored, visible and correctable.
Where a model is used
- Prefilling your company form from a website, a PDF or a paragraph, when you ask for it. You see and edit every field before screening. Model: gemini-3.8-flash.
- Enriching records offline. When a funder publishes eligibility only as prose, the model extracts beneficiary types, sectors, regions, size and age limits and consortium requirement into a separate table, once per record. The matcher uses those fields only when the funder published none.
Where no model is used
- Collecting and normalising the ledger: code only, from official APIs and open data.
- Search, filters, deadlines, history.
- Screening verdicts, scores, readiness gaps and negative memos: fixed rules, listed on the result page.
What can go wrong
A model can misread a document or a call: wrong country, wrong sector code, a condition missed. Prefilled fields are suggestions until you confirm them. Enriched fields are marked as such and can be wrong; the official call text always wins. Report a problem from any record page.
Your data
Model calls go through Cloudflare AI Gateway to Google. Your inputs are not used to train any model. Your form, your uploads and your results are stored privately, never published, never indexed, and you can export a result as JSON from its page.
Providers
Google Gemini through Cloudflare AI Gateway. The provider is one file in the open source code and can be swapped; the screening does not change when it is.