The EDC is live today — see what ships now
Trialion

Clinical-trial intelligence

Clinical trialsstill run ontranscription.

A protocol becomes a form by hand. A lab report becomes data by hand. A discrepancy becomes a query by hand, weeks after entry. Trialion is building the intelligence layer that reads the documents, drafts the work, and stops for a qualified signature. It is live today, as an AI-native EDC.

See it on your protocolHow the engine works
EDC & eCRF
Live
AI Designer
Live
AI Capture
Live
Signatures & SDTM export
Live
TTrialion Engine · TRL-0412 · Visit 2Demo
01 / Read3 docs
  • PROTOCOL3 pp
    §7.2 Vital signs
  • LAB PDF2 pp
    Central lab · SUBJ-00482
  • SOURCE1 pp
    Scanned note · Site 04
02 / Draft3 drafts
  • FORMfrom §7.2
    14 fields · 6 edit checks
  • VALUESfrom lab PDF
    4 extracted · conf ≥ 0.94
  • QUERYSite 04
    CHK-003 · systolic > range
03 / Sign2 / 3
  • DM09:14
    Form v1.2 approved
  • CRC09:31
    4 of 4 values confirmed
  • DMopen
    Query awaiting review
One data modelOne audit trail21 CFR Part 11

Illustrative — not a live study

Designed for & aligned withCDSCO Schedule YIndian GCPICMR (2017)DPDP Act 2023CTRI registrationICH E6(R3)21 CFR Part 11CDISC ODM / SDTM

The problem

Most of a trial's data work is retyping.

01 /

The protocol is rebuilt as forms, by hand.

A data manager reads the protocol and builds the CRFs, the field types and the edit checks from it. Four to eight weeks on most studies, and a version of the same work again at every amendment.

02 /

Source is typed in, then checked against itself.

Lab reports arrive as PDFs. Remote sites still work on paper. A coordinator transcribes each page into the EDC, and a monitor later verifies the transcription against the page it came from.

03 /

Discrepancies surface weeks after entry.

An out-of-range value sits in the database until a monitoring pass finds it. By then the subject has had two more visits and the query is about something nobody at the site remembers.

04 /

Four systems, reconciled in a spreadsheet.

Recruitment in one vendor, randomisation in another, dispensing in a third, the eCRF in a fourth. The reconciliation is a spreadsheet, and the spreadsheet is where the errors live.

Every one of these looks like an operations problem. Every one of them is a reading problem.

The engine

One engine under the EDC.

The same pipeline reads a protocol, a lab report or a scanned source page, drafts the structured work that follows from it, and holds that draft until a qualified person signs. The EDC is built on it.

01 /

Live

Read

Protocol sections, central-lab PDFs, scanned source pages. The engine reads the documents a trial is actually made of, and keeps a pointer back to the paragraph or the place on the page every value came from.

02 /

Live

Draft

Forms and edit checks from the protocol. Values from the report, each with a confidence. A query from the discrepancy, with subject, site and visit attached. Every one is a proposal, written down with its reason.

03 /

Live

Sign

Nothing counts until a qualified person accepts it. The reviewer, the timestamp, the value before and the value after land on the same audit trail as the data, under 21 CFR Part 11 signatures.

Walk through it, step by step

How it works

Four steps. Two of them are a person.

The AI Designer drafts the form. AI Capture reads the source documents. Both stop at a person — nothing is saved, and nothing reaches a monitor or a regulator, until someone qualified has put their name to it.

Step 01 / Draft — AI Designer

Give it the protocol. Get the form back.

Point it at a protocol section and it drafts the fields, the field types and the edit checks that follow from them, with the paragraph each one came from still attached.

Then it stops. A draft is a proposal, not a study — nothing reaches a site until a data manager has read it and approved it.

  • · Fields and types drafted from the protocol text
  • · Edit checks proposed with the rule spelled out
  • · Every draft traceable to the section it came from

Or build it yourself

Change anything, without writing code.

Drag field types onto the form, set the conditional logic and add edit checks. The AI is optional — a study can be built this way from an empty form, and every draft it proposes can be taken apart the same way.

  • · Conditional logic without programming
  • · Edit checks at the point of entry
  • · Amendments deploy without a rebuild

Step 02 / Capture — AI Capture

Upload the lab report. Skip the typing.

A central lab PDF, a scanned page from a site that still works on paper — AI Capture reads it and lifts the values into the form, each one carrying how sure it is and a link back to the place on the page it came from.

Nothing is saved on the model’s word. A person checks the value against the source and confirms it, and that confirmation is what lands on the audit trail.

  • · Sites keep working the way they already do
  • · Every value traceable to its place on the page
  • · Confidence shown, so a low one gets a second look

Step 03 / Check

Catch it. Rank it. Draft the query.

Edit checks fire as data lands, not in a monitoring pass weeks later. The discrepancy arrives with subject, site and visit attached, and the site query is drafted and logged to the audit trail.

It then waits. Nothing leaves the platform until a data manager reads it and decides it should.

  1. 1Detect
  2. 2Prioritise
  3. 3Draft
Check #003Running1 issue found

Consent date precedes first study procedure

Verify that consent is signed before any visit or drug exposure.

Consent date is missing
SUBJ-00214 · Site 04 · Visit 1
Draft site query

Step 04 / Sign

The part that stays human.

Every draft the platform produces is a proposal until someone qualified accepts it. The reviewer, the timestamp, the value before and the value after are written to the same audit trail as the data itself.

  • · Attribution and reason-for-change on every edit
  • · Electronic signatures under 21 CFR Part 11
  • · One immutable trail, queryable at inspection

The product

An EDC on one data model.

Forms, edit checks, queries, signatures and exports read and write the same records under one audit trail. There is nothing to reconcile inside the EDC, because there is only one system. Everything else — your CTMS, eTMF, eConsent, RTSM and safety database — connects once, at the edge.

01 /

Live

EDC & eCRF

Design forms and capture clean data, fast.

  • Visual eCRF designer with conditional logic and derivations
  • Real-time edit checks and automated queries at point of entry
  • Protocol amendments and mid-study changes without rebuilds
Module detail

02 /

Live

e-Signatures, Freeze & Lock

Regulated-grade signatures, freeze, and lock.

  • 21 CFR Part 11 electronic signatures
  • Data freeze and database lock with attribution
  • Immutable, queryable audit trail
Module detail

03 /

Live

Reports, Export & SDTM

Reports, SDTM-ready export, and database lock.

  • Reports and analytics across the trial data model
  • SDTM-ready, submission-grade exports
  • Scheduled exports to warehouses and BI tools
Module detail

Everything above the line is the EDC, live today. Everything below it is a system you already run, connected by integration.

External integration will always be a requirement. What changes is that it happens once, at the boundary — not repeatedly between internal modules.

Every supported integrationEvery module

Compliance

Compliance sits in the schema, not on top of it.

Audit trails and electronic signatures are properties of the data model, so no module can be configured out of them. Designed for the frameworks below, in the order an Indian sponsor asks about them.

The compliance detail
  • CDSCO Schedule Y
  • Indian GCP
  • ICMR (2017)
  • DPDP Act 2023
  • CTRI registration
  • ICH E6(R3)
  • 21 CFR Part 11
  • CDISC ODM / SDTM

Specification

Numbers we can stand behind.

30 minutes
To see it running on your own protocol
No scoping call, no quote and no purchase order first.
10 minutes
To stand a study build up end to end
The eCRF itself is instant — the AI Designer drafts it from your protocol.
No code
To build your forms, checks and visit schedule
Doing it by hand takes most teams 4–8 weeks.
One
Data model behind every module
Enter data once and it is everywhere. Nothing to match up later.
Zero
Extra fees per form, per query or per export
The price is the same whether you collect 10 records or 10,000.
Most systems we looked at were built for a trial that does not look like ours. Trialion starts from Schedule Y, and from what a coordinator at a remote site can realistically do on the day. The audit trail is the part that holds up when an inspector starts asking.
SSSushant SinghFounder, Clinical Veda

Before a build

The questions we get asked first.

Answered as we would answer them on a call, including the ones where the answer is not yet.

01What environments do I get, and can test data reach production?

Today a study moves draft → UAT → live, and configuration is promoted between them. A separate development environment and an independent QC stage, with test subject data isolated from production, is the next piece of architecture we are building. If you are planning a build, ask us where it stands on the day.

02Show me a mid-study amendment with subjects already enrolled.

Amendments version the study configuration and deploy without a rebuild. What we will not claim yet is a complete change-control workflow across separate environments, with a documented impact on already-enrolled subjects. Bring the scenario to the call and we will walk it honestly rather than demo around it.

03How is my study validated, separately from your platform?

They are two different things and we keep them separate. Platform validation is ours: the system is architected to GAMP 5 and the formal exercise runs with an external auditor. Study validation is proving that your CRFs, edit checks, derivations and visit structure behave to your specification — today that happens in UAT, and a dedicated QC stage with retained evidence is in build.

04Can we configure roles beyond PM, DM, CRA, CRC and PI?

Yes. Roles and their permissions are configurable rather than fixed, at organisation, study and site level, because no two CROs use the same titles. Field- and form-level restriction is the next level of granularity and is not there yet.

05What happens when a required field genuinely cannot be filled?

Today it raises a query, and that is wrong — a deliberate blank is not a discrepancy. Standardised missing-data reasons (not done, not applicable, unknown, not available) recorded with a comment on the audit trail are a change we are making, because statistics needs to tell an omission from a fact.

06Do we have to use the AI?

No. A study can be built entirely by hand, and nothing the model drafts takes effect until a person accepts it. The AI writes a proposal; a qualified reviewer decides whether it becomes the study.

07What is retained about an AI-drafted form?

The generated output, every human change made to it afterwards, and who approved it are on the audit trail. Retaining the model version and the prompt alongside them is a change in progress, so an inspector can see what was proposed as well as what was accepted.

08Do you support blinded studies?

Not today. Trialion is built for open-label work, including bioequivalence and Phase I–IV designs that randomise without blinding. Blinding touches randomisation, drug supply, unblinded pharmacy roles and emergency unblinding at once, and we would rather say no now than ship something that leaks a treatment assignment.

Not every protocolfits in a template.

Bring the one that does not. Thirty minutes, your own protocol, and you will see it running rather than described. Sandbox in 24 hours.

See it on your protocolSee pricing