WanderCode logoWanderCode

Grants AI — AI-Assisted Grant Research, Drafting, and Reporting

We turned a fragile grants prototype into a hardened, AI-assisted platform that researches past winners, drafts applications, and produces live-metric funder reports.

Grants AI — AI-Assisted Grant Research, Drafting, and Reporting

Overview

We partnered with Grants AI to harden an existing grants implementation and extend it into a complete, AI-assisted workflow spanning past-winner research, application drafting, live-metric reporting, lifecycle email handling, and video and webinar knowledge ingestion. We delivered a tenant-safe, grounded, and auditable system that grant teams can trust end to end.

Challenges

  • The existing grants implementation could not be safely shipped, carrying a backlog of pre-merge blockers around permissions, tenant schema context, AI stubs, and broken client imports.
  • Access control gaps and incorrect tenant schema-context handling created real risk that one tenant's grant data could leak into another's.
  • A vector-extension database migration was placed incorrectly, threatening to disrupt other tenants on the same cluster.
  • AI research could silently accept ungrounded or unsupported past-winner results, eroding trust in the data filling each grant record.
  • External research pulled in URLs with no validation, exposing the system to local files, metadata IP ranges, and internal network targets.
  • Grant drafting was one-off and disposable, with no persisted draft records, status tracking, or way to regenerate a single weak answer.
  • Reporting lacked a safe path to live program metrics, and there was a risk of accidental writes back into upstream modules that should stay read-only.
  • Grant lifecycle work was untracked, with no email classification, deadline alerts, auto-archiving, or detection of missing award letters.
  • Valuable funder knowledge locked inside videos and webinars was never captured or made available during drafting and reporting.
  • A stray PDF dependency added unnecessary surface area and risk to the scoped work.

Solution highlights

  • Remediated every pre-merge blocker, fixing permissions, tenant schema context, AI stubs, the client import path, vector migration placement, and removing the PyPDF2 dependency.
  • Built real past-winner extraction using structured AI output so research reliably populates database fields.
  • Added grounding verification with hard-fail behavior, so unsupported research results are rejected rather than silently downgraded.
  • Introduced URL validation that blocks unsafe targets, including local files, metadata IP ranges, and internal network ranges.
  • Shipped a grant application draft model with migrations, serializers, and a status state machine, turning drafting into persisted, trackable records.
  • Built an AI drafting service using retrieval, prompt assembly, and per-question regeneration, paired with a frontend grant-drafting editor page.
  • Created a grant report model with funder templates and a cross-plugin metrics read interface so reports substitute live program metrics without write access to upstream modules.
  • Added grant lifecycle email ingestion through a tenant mailbox, with an email classifier, auto-archive rules, missing-award-letter flagging, deadline notifier, in-app alerts, and an email digest.
  • Surfaced a correspondence timeline on the grant page so every lifecycle event is visible in context.
  • Delivered video and webinar ingestion via URL input, transcript generation, chunking, embedding, and retrieval wiring, plus a video library panel in the frontend.
  • Delivered REST endpoints, Celery tasks, and a constrained structured-target web-navigation proof of concept for research.

Outcomes

  • Grant teams can now ship the platform with confidence, knowing the pre-merge blockers and tenant-safety risks have been resolved.
  • Tenant data stays isolated, with corrected schema-context handling and vector migrations moved into the correct cluster-safe path.
  • Past-winner research lands as reliable, structured data in the database instead of unverified free text.
  • Every research result is grounded and verifiable, with unsupported outputs failing fast rather than slipping through.
  • External research is safer by design, with unsafe and internal URLs rejected before they can be followed.
  • Drafters work from persisted draft records with status tracking and can regenerate any single answer without redoing the whole application.
  • Reports pull live program metrics through a read-only interface, keeping upstream modules protected from accidental writes.
  • Grant lifecycle management is automated, so correspondence, deadlines, alerts, missing award letters, and archival are tracked without manual chasing.
  • Funder knowledge from videos and webinars becomes searchable context that strengthens both drafting and reporting.

Gallery

Grants AI — AI-Assisted Grant Research, Drafting, and Reporting screenshot 1
Grants AI — AI-Assisted Grant Research, Drafting, and Reporting screenshot 2
Grants AI — AI-Assisted Grant Research, Drafting, and Reporting screenshot 3