Coreagentic Grant Proposal System
Read a funder’s solicitation the way a reviewer will, decide whether it fits before anyone opens a blank page, and check each section of the draft against the criteria it will be scored on. I built it with a two-person grants team, and it is designed so that team makes every decision.
What it does
Extracts the rubric
Each solicitation gets one record. The system reads the RFP or NOFO and pulls the funder's review criteria and published scoring weights into that record. A federal juvenile-justice funder, for example, publishes a 10/45/25/10/10 split across its criteria. Knowing that before the first sentence decides which section gets the week and which gets the afternoon.
Scores fit
Before anyone commits staff time, the system judges how well the opportunity fits your organization. It works from your own mission documents, past proposals, and the house rules your staff write down: service gaps, team skills, restrictions. Every reason for and against is a verbatim quote, checked against its source, so you can open the document and read the paragraph it came from.
Scores the draft
Once a draft narrative exists, the system scores its strength against the extracted criteria, section by section, and offers concrete suggestions. Run it before internal review and again before submission.
Design rules
Built to be trusted in front of a funder
Bands, not percentages
Fit and strength come back as bands. A model that reports 73 percent is claiming a precision it does not have. A band tells your team what it needs to know and nothing it cannot back up.
The math lives in code
The overall score is computed in code from the component findings. The model reads and judges. It does not do the arithmetic.
Unknown caps the result at Fair
Anything the system cannot judge is marked Unknown, and any Unknown caps the overall result at Fair, however strong the rest looks. The Unknown also points at the fact about your organization nobody has written down yet.
Every reason is a quote
Reasons come from the solicitation or your own files, word for word, and are checked against the source before they are shown. Nothing is paraphrased.
Scores go stale on their own
When the criteria or the draft change, the old score is marked stale, so last cycle's band is never mistaken for this one's.
Advises only
The system never approves, returns or blocks a proposal. Your reviewers keep every decision. Humans decide, agents execute.
How a grants team uses it
Triage
Run each incoming solicitation for fit before anyone commits staff time to an application. When the band is Poor, the decision to pass is short, and it is saved with the quotes attached for the next time that funder publishes.
Write to the rubric
Pull each funder's review criteria into one place, so the draft is written to the rubric from the first page.
Check before review
Run draft sections through the strength check before internal review and before submission.
Track
Follow each application through its review stages, with email alerts and the application files kept in SharePoint.
Where it stands
Built with a real grants team
I built the Grant Proposal System with a children’s advocacy nonprofit whose grants team is two people. Before the pilot, we ran it on eight real solicitations and a 37-file grant store, then presented it to their board with a request to deploy it into their own Azure tenant.
The live pilot on a real grant starts in November 2026, and the results go back to their board in January 2027. Two more organizations are evaluating it now.
I can’t share how those first eight solicitations scored. That is the customer’s information. What I can show you is how the scoring works, on an example, with every quote behind it.
Questions grant teams ask
Does it decide which grants we go after?
No. It puts a band and the quotes behind it in front of your team. Your people decide whether to apply, what to write and when to submit. The system never approves, returns or blocks a proposal.
What does it cost?
One fixed annual fee, quoted before you commit. No per-proposal or per-seat charges.
What does it need from us?
Your mission documents, past proposals, and a short set of house rules your staff write: the service gaps you fill, the skills on your team, and the restrictions you work under. The nonprofit it was built with started from 37 files. The more of your organization is written down, the fewer criteria come back Unknown.
Where do our documents and drafts live?
The system runs as a web app over an indexed set of your organization's documents. Today it uses OpenAI's models. The deployment we are moving to runs Azure OpenAI inside the organization's own Azure tenant, so funder documents and drafts stay in that organization's cloud.
What kinds of solicitations does it read?
Funder solicitations such as RFPs and federal NOFOs. Each one gets its own record holding the extracted criteria, the fit score and, later, the draft scores.
See a scored solicitation
I’ll walk you through a scored example, the quotes behind each band, and what it would take to point the system at your own documents.
Book a walkthrough