An R1 research university pre-award office assembling NIH R01 and NSF proposal packets in the last week before a deadline, section by section, with boilerplate buried in a shared drive. We built the assembly layer: LlamaParse on typed proposal sections, Azure AI Document Intelligence on signed institutional documents, every packet checked against the funder template with section ownership named and missing items surfaced for pre-award.
| Proposal | PI | Sponsor | Sections | Deadline | Status |
|---|---|---|---|---|---|
| PR-24871 | Acharya, R. (Bioengineering) | NIH R01 | 13 of 14 | May 5 | on track |
| PR-24872 | Okonkwo, T. (Public Health) | NIH R21 | 11 of 13 | May 7 | PI revising |
| PR-24873 | Larsen, M. (Civil Eng.) | NSF CAREER | 14 of 14 | May 12 | on track |
| PR-24874 | Brennan, J. (Education) | NSF EHR | 10 of 13 | May 8 | IRB hold |
| PR-24875 | Patel, S. (Biology) | NIH R01 | 12 of 14 | May 14 | on track |
| PR-24876 | Yamamoto, K. (Chemistry) | DOE EERE | 9 of 12 | May 19 | budget rev. |
| PR-24877 | Diaz, L. (Public Health) | NIH K23 | 13 of 14 | May 21 | on track |
| PR-24878 | Whitfield, A. (Mech. Eng.) | NSF CMMI | 14 of 14 | May 26 | on track |
At a glance
One pre-award office, two funders in scope, one submission system per funder. The template check and section ownership were the pieces that closed the deadline week.
The engagement
The stack
ISO 27001 · ISO 9001 · FERPA scope · DPA and NDA signed at kickoff.
Before, the pre-award desk
Pre-award assembled every packet manually. Section ownership sat in a shared spreadsheet, boilerplate lived on a share drive, and the deadline week compressed the review into hours.
NIH R01 required 11 primary sections plus attachments. NSF required a different mix. Pre-award tracked the list in a spreadsheet per proposal, ticking items as PIs sent drafts. On a clean proposal, the list closed a week out. On a typical proposal, the list closed the day before.
Pre-build baseline: template completeness cadence varied, with close often arriving in the final 48 hours.
The PI owned the science sections. The department administrator owned the budget. Pre-award owned the institutional attachments. Who owed what surfaced in email chains, not on the proposal record. Follow-ups took days to resolve.
Pre-build baseline: ownership-to-section mapping reconstructed per proposal, not recorded on the packet.
F&A rates, facilities statements, compliance assurances: all on a shared drive that the research office updated when they remembered. PIs pulled old copies. Pre-award caught most of the drift, but not all.
Pre-build baseline: approximately 12% of packets required boilerplate correction in the final review.
What we built
The pipeline follows the same five stages we run on every assembly engagement. The funder template library and the section ownership map are tuned against this university's two funders in scope.
PI email polled on a 10-minute cadence, Cayuse draft folder listener, Kuali Research working-copy webhook, institutional document drop to SFTP. Every proposal assigned a single proposal ID.
Each document tagged to a funder template slot. Science sections, budget justification, and institutional attachments routed separately but linked by proposal ID. Classification confidence below 0.90 holds the document.
PI name, section title, F&A rate used, effective dates, signature presence. LlamaParse on typed proposals, Azure AI Document Intelligence on signed institutional documents.
Each proposal checked against the funder template. Sections mapped to named owners. Institutional boilerplate cross-checked against the current version. Below 0.90 confidence, the packet holds for pre-award review.
Clean packets posted to Kuali Research or Cayuse with section ownership named on the packet. Missing items surface with the owner listed in plain English. Pre-award reviews completeness, not the checklist.
After, the numbers the pre-award office signs off
Same pre-award staff, same two funders, same deadline cadence. The pipeline assembled every packet against the funder template, named the owner per section, and caught outdated boilerplate before pre-award review. Deadline weeks stopped being the bottleneck.
Pre-award still own the submission. They still review every packet before it leaves the institution. The difference is that the open-items list sits on the proposal, the owner is named, and boilerplate drift gets caught before the final review.
From the desk
The last 48 hours used to be where our errors lived. Section ownership on the packet pushed the review forward, which is where it belongs.
Pre-award operations leadR1 research university, West Coast
Handover
The engagement ends at a clean handover. The pre-award office runs the pipeline; Hexaa stays on call for a fixed retention period, then steps back.
Related cases
Each links to a named client, a named document, and the system the clean data lands in. We publish only what the client signed off to publish.
Incoming transcripts normalised to the course catalogue across 140 feeder institutions, with under-threshold equivalencies flagged for evaluator review.
→State and local government · 2026Mid-sized city · permit intake, one queue per typeBuilding, zoning, and right-of-way permits classified into one queue per type with attachment checks at intake before the clerk opens the file.
→Construction · 2026General contractor · AIA G702 and G703 packet assemblyAIA pay application packets assembled with supporting attachments named per line. Procore as the system of record.
→Free 30-minute call
You'll leave with a clear next step.
An NIH R01 packet arrives with 9 of 11 sections drafted. The pipeline checks the template, names the missing specific-aims section and the outdated F&A boilerplate, and routes each back to the owner listed on the packet. Pre-award opens a work list, not a chase.
