AI-assisted evidence review

Review applications faster.Decide with confidence.

ReviewOS uses AI to identify missing evidence, mismatches, and next steps so your team can make faster, fairer decisions — with humans in control.

Get started See how it works

No sign-up required · Guided tour included

Finds what's missingAI flags gaps and suggests next steps.
Checks for mismatchesHighlights inconsistencies across documents.
You stay in controlAI provides insights. You make the call.
3 evidence itemsRegistration, activity plan, signoff — completion is provided / 3.
AI assists, humans decideReview Ready means complete + consistent. Never an approval.
Deterministic safety netRule-based checks veto an LLM verdict when evidence fails.

What it does

Everything a reviewer needs, nothing they don't

Finds what's missing

AI flags gaps across the three required evidence items and suggests next steps.

Checks for mismatches

Highlights inconsistencies across documents — e.g. organisation names that don't line up.

Drafts the applicant request

Generates a polite, specific message asking for exactly what's outstanding.

How it works

From flagged to Review Ready

Follow the exact path a reviewer takes: open a flagged application, run analysis, draft the applicant message, simulate the fix, and re-analyze.

Get startedView an example
  1. Open a flagged applicationStart from the application queue — APP-101 or APP-102.
  2. Inspect the evidence checklistThree required items: registration record, activity plan, responsible-person signoff.
  3. Run AI analysisDeterministic checks run first; an LLM summary is layered on when a key is configured.
  4. Generate the applicant requestOne click drafts the message for the missing or inconsistent evidence.
  5. Simulate the fix, then re-analyzeProvide the missing item and re-run analysis to reach Review Ready.

Evidence model

Three documents. Four statuses. Zero ambiguity.

Required evidence

  • Registration recordrequired
  • Activity planrequired
  • Responsible-person signoffrequired

Completion is simply provided / 3 — 33%, 67% or 100%. Mismatches never lower completion; they block readiness instead.

Review statuses

  • Analysis RequiredNot yet analyzed — or evidence changed since analysis.
  • Missing EvidenceA required item is missing. Not ready for review.
  • Needs ClarificationEvidence present but inconsistent. Human clarification needed.
  • Review ReadyComplete and consistent. Ready for a human decision.

Product tour

Three applications, three outcomes

View all
APP-101Learning Workshop AVocational pilot · 67% complete · Signoff missingMissing EvidenceAPP-102Community Workshop BTrainer development · 100% complete · Org-name mismatchNeeds ClarificationAPP-103Northgate Skills CollectiveVocational pilot · 100% complete · Complete + consistentReview Ready

MCP server

Bring your own LLM — 4 tools, 5-minute setup

Connect Claude Desktop, MCP Inspector, or any MCP client to list, inspect, analyze, and draft requests — live API first, with a deterministic fallback so reviews keep working.

MCP setup guide

See a full review cycle in minutes

Open the application queue, pick a flagged application, and walk the list → analyze → request → fix → Review Ready path.

Get started See an example case