GreenLight

AI for clinical trial operations

Your trial's back office, made visible

AI operations for clinical trial sponsors

Your trial runs on paperwork.
GreenLight reads every page.

1572s, IRB letters, contracts, invoices. GreenLight extracts what matters, checks it against the contract and the source record, and flags every mismatch with the page and line cited. Your team makes the calls.

Request accessSee how it works
Backed byLeapYearLeapYearAfore CapitalAfore Capital
Site 022 · Source packetReceived 09:41

FDA 1572 · CTA schedule B · Invoice #1042

Principal investigatorDr. Sofia BrandtCV match · verified
Medical licenseMD-44921Active · exp 03/2027
MRI · visit 4$1,850≠ $1,200 contracted cap

Invoice line 7 vs CTA schedule B · both pages cited

SHA-256 sealed3 cited · 1 flagged · awaiting review
Scroll
01Reads every page the moment it arrives
02Extracts the facts, cited to the line
03Checks them against the contract, flags what doesn’t match
04Then waits for your team to make the call
Built by operators from
  • Harvard Medical School · MIT HST
  • Columbia University · Vagelos College of Physicians and Surgeons
  • Howard Hughes Medical Institute
  • Emory Winship Cancer Institute
  • MD Anderson Cancer Center
  • Harvard Medical School · MIT HST
  • Columbia University · Vagelos College of Physicians and Surgeons
  • Howard Hughes Medical Institute
  • Emory Winship Cancer Institute
  • MD Anderson Cancer Center

Institution names describe prior team experience only. No endorsement or affiliation implied. Prior experience in clinical research operations, regulatory evidence, trial finance, and multi-site study execution.

How it works
4 steps
FDA 1572 · receivedFinding F-208 · evidenceSite 022 · activationApproval · audit trailGreenLight

FDA 1572

North Valley Research · p. 1

PI name

Dr. Sofia Brandt

License

MD-44921 · active

Protocol

v4.0 · signed

Site

North Valley Research

Invoice #1042 · line 7

MRI · $1,850

≠

Executed budget · CTA

MRI cap · $1,200

Blocking · over contracted rate

$650 above the contracted rate

Evidence · invoice line 7 vs CTA schedule B · both pages cited

Nothing is auto-fixed. This waits for your team.

Site 022 · activation gates

IRB approvalCleared
GCP trainingComplete
SIV readinessScheduled
Contracts executedCleared

Ready for first patient in

Finding F-208

Over contracted rate · $650

DismissApprove Done

Audit trail · append-only

14:02:11F-208 approved · j.chen, Sponsor Ops
14:02:11sha256 9f3a…c41d · version sealed
Every action, human or AI, is recorded permanently.

01Read

Every page is read the moment it arrives.

Sites and CROs submit through a tokenized link. No login, no portal to learn. AI reads 1572s, licenses, ICFs, contracts, and invoices the moment they land, pulling names, dates, protocol versions, and credential numbers straight off the page in under a minute. Nothing is retyped by hand.

02Check

What doesn’t match gets flagged, with proof.

Every extracted fact is checked against four sources: the executed CTA budget, prior billing history, EDC visit data, and the signed protocol. Duplicate charges, over-rate billing, expired credentials, and version mismatches all surface the same way, with the exact page and line cited.

03See

Every gate, cleared with evidence.

IRB approval, GCP training, SIV readiness, executed contracts: each gate closes only when the evidence behind it checks out, not just when a document lands. A gate that can't close yet tells you exactly what's missing and who's holding it up.

04Approve

A person makes every consequential call.

AI never resolves a finding and never moves money. Your team approves or dismisses each one. Approved items export as an audit-ready payable packet or activation package. Every decision is time-stamped, attributed to a named user, and hash-sealed so it can't be quietly edited later.

The platform

Two surfaces.
One standard of proof.

GreenLight runs the document- and money-heavy back office of a trial across two surfaces, built on the same evidence-first spine: extract the facts, flag what doesn't match, cite the source, wait for a human.

Site activation

Everything between choosing a site and treating a patient, collected, checked, and tracked in one place.

  • Tokenized upload portalsSites submit documents through a link. No accounts, no logins.
  • AI validationIdentity, version, and credential checks on every page.
  • Findings queue & activation blueprintA per-site plan; only real problems reach your team.
  • IRB workspacePacket assembly and approval-expiry detection.
  • IND / FDA workspaceDrafting support and eCTD hand-off.
  • Contracts & e-signatureAI-suggested redlines, tracked to execution.
  • SIV readiness & multi-countrySite-initiation checks, EU & IVDR activation abroad.

Budget operations

Every invoice line reconciled against the executed budget, prior billing, and patient-visit evidence before anyone pays it.

  • CTA reconciliationEach line checked against the executed trial budget.
  • Duplicate billingCaught against the full prior billing history.
  • Over-contracted-rate chargesFlagged with the contract page cited.
  • Quantity over cap & screen-fail billingPer-protocol limits enforced on every line.
  • Visit verificationInvoiced visits matched to EDC patient records.
  • Audit-ready payable packetsExported with evidence attached. Never touches payment rails.
Agents

An agentic layer that does the chasing.

GreenLight ships with agents for practically everything sponsors and CROs do by hand: answering questions, filing documents, chasing sites, working the inbox. Every action comes from an audited capability registry, and every consequential one waits for a person to say yes.

  • In-app assistant

    Ask about any trial, site, or invoice in plain language. It answers from your own tenant-scoped data and drafts next steps; a person confirms every change.

  • Capability registry

    Every action any agent can take comes from one fixed, audited, continuously validated list. Nothing is improvised.

  • Chase system

    Predicts which sites are about to stall and escalates the exact person holding up a document, on a cadence, automatically.

  • Mailbox autopilot

    Files documents straight from your Gmail or Microsoft inbox, matches them to the right site, and chases whatever is missing.

  • External agents · MCP

    Your own AI tools call GreenLight capabilities through the same audited pathway a person uses.

Trust

Built to survive an audit.

Sponsors carry the regulatory risk, so GreenLight is built the way an inspector would want it: nothing quietly changed, nothing unattributed, nothing automatic where it matters.

  1. 01

    AI is assistive, never authoritative

    It reads, matches, and flags. It never resolves a finding, never auto-fixes a document, and never moves money. A person makes every consequential call.

  2. 02

    Immutable, versioned records

    Every document and invoice version is kept, time-stamped, and content-hashed. You can always see the current version, and every version before it.

  3. 03

    A tamper-evident audit trail

    Every action, human or AI, is recorded permanently with who, what, and when, built to the standard of a 21 CFR Part 11 inspection.

  4. 04

    Your data stays yours

    Each sponsor’s trials are isolated from every other tenant’s, enforced at the database layer, not by convention.

Questions

Before you ask your CRO.

  • No. AI extracts, matches, and flags, but it never resolves a finding and never touches payment rails. Every consequential action waits for a person on your team to accept or dismiss it.

GreenLight

Run your next trial
on evidence.

Early access is open to clinical trial sponsors, and the CRO teams running programs on their behalf. Tell us about your studies; we'll respond within 48 hours.

AI assists. Your team approves. Nothing moves money on its own.