Every financial-aid decision has a rule, a citation, and an auditor who may ask. Eli works the exception queue your office already runs and cites the rule behind every number.
Offices can't hire their way out of exception work.
The IRS data exchange now imports most FAFSA tax data as verified. What remains is manual:
Most verification tools collect and route documents, then stop. Eli works the queue.
Supported calculations run on unit-tested engines keyed to versioned rules: the $25 correction trigger, all-or-nothing submission, IRS Per Computer precedence, the SAI recomputation, SAP standards (34 CFR 668.34), the R2T4 return order (668.22), loan proration and caps. Directional SAI and Pell previews are labeled apart from final federal-table results. Unsupported cases stop for review.
Models extract, classify, summarize, explain, and draft, each inside a typed task. Every extracted value carries a source and a confidence. Calculations stay in tested code. Anything that turns on policy or judgment goes to staff.
Identity rejects, V4/V5 outcomes, conflicts, SAP and PJ decisions, R2T4, and anything low-confidence route to a reviewer with the case history. Eli drafts. A person decides. The demo stages simulated or file-based action only; any production SIS or federal write needs a scoped integration and named authorization.
Every action lands on a replayable record: what Eli read, what it recommended and why, the rule, and who made the call.
Every aid workflow runs the same ten stages. Calculate is number five. The case closes when the posting is acknowledged, the student is told, and the money ties out. Eli tracks all ten.
A professional-judgment case. Eli reads the documents and recomputes the Student Aid Index. The card shows its confidence and its reasoning. An administrator decides.
Recompute SAI with documented income loss, then route for a professional-judgment decision.
Eli grades itself against labeled documents. Most recent run: 2026-06-09, 21 synthetic fixtures, 52 fields.
Synthetic fixtures: clean and OCR-degraded renders, not scanned IRS originals. A scanned benchmark is next. Rule calculations are unit-tested separately. Production quality has to be measured on contracted data.
Three institution types, all synthetic. Sign in as Eli Operations for the cross-institution view.
Open the demoSynthetic data only. FERPA role, FTI scope, and Title IV third-party-servicer status depend on the contracted work and need institution and counsel review.