Synthetic data · No PHI · Demo environment · Not for clinical use
Synthetic data · No PHI · Demo environment · Not for clinical use
Clinical trial operations · Site + Sponsor

The AI-native operating system for clinical trials, site to sponsor, protocol to close-out.

TrialFlow's agents draft the protocols, trial documents, eligibility checks, safety reports, and billing work every trial runs on. A clinician reviews and signs each one. Every patient, visit, and action tracked to a permanent audit trail — replacing Excel sheets and manual record-keeping.

One product for site and sponsor · One data model · One audit trail
app.trialflow.ai / trial / IMMPACT-001
Amendment v4 → v5 · agent runreplay
AgentAmendment ingested · diffed against current protocol
AgentRegenerating 2 of 47 affected documents · 45 held at approved version
QCFlag · §7.4.1 tocilizumab — "max 800 mg" dose cap missing
HumanClinical Research Analyst (a named human) reviewed · resolved
SignInvestigator e-signs · 21 CFR Part 11 · hash sealed
GateAmended protocol + consent → ethics board (IRB) · re-consent on approval
Built on the standards cancer centers and sponsors already run on
Anthropic ClaudeEpic FHIRMedplumHIPAA21 CFR Part 11SOC 2
One product for site and sponsor. One data model. One audit trail.
Built to lift the operational burden off every trial role
CRCPICRAMedical MonitorRegulatory AffairsQM
The missing layer

Oncology trials get delayed, over-budget, and stalled before they ever produce science.

7:15 AM47 unread emails across three sponsors and the ethics board.
7:32 AMEight trial trackers open in Excel, cross-checked against the weekend's lab results.
7:48 AMAt the sponsor's office in Cambridge, a CRA opens 14 site monitoring reports. Three sites missed protocol deviations last cycle. The medical monitor still hasn't signed the last DSUR draft.
8:05 AMA cell-therapy patient was admitted Saturday. No one was told. The 24-hour safety-reporting clock has been running since Saturday morning. It's now Monday.
8:15 AMA protocol amendment came through Friday. Nobody's opened the redline yet. Which of the 47 trial documents actually need to change?
8:41 AMA patient arrives for treatment. A consent-form update from four weeks ago was never scheduled for re-signing.
9:00 AMAnd the day hasn't started.

None of this is for lack of effort. It's general-purpose software (Excel, email, shared drives) doing a job it was never built for. That gap is where trials delay, deviate, and stall.

76%
of trials now need a major amendment — up from 57% in 2015
Getz et al., TIRS 2024
$55,716
direct cost of a single day of Phase III delay
Tufts CSDD, 2024
167 days
to open a trial a site's own doctors design — vs a 90-day target
Ratnayake et al., 2025
86%
of research coordinators report work-related distress
Florence Healthcare, 2023
Most software just stores. This one does the work.

Not a database. A system that does the operational work. And signs it.

Trial software today either stores the records or holds the files. Every system tracks something else: which trials exist, what data the sponsor needs, what regulators require, which patient might be eligible. None of them do the actual work. TrialFlow is built around the trial and the patient. Its agents draft the documents. Grade the side effects. Reconcile the billing. Prepare the filings. A human reviews and signs each one before it goes anywhere.

Trial databases (CTMS) — OnCore, Veeva
Store what happened. TrialFlow generates the records in the first place.
Document binders — Florence, Complion
Hold your files. TrialFlow writes the documents that fill them, from the protocol.
Patient-finding AI — Deep-6, point-of-care tools
Flag a candidate. TrialFlow does everything after the name is flagged.
How it runs

Agents draft. Humans approve. Everything is signed.

AI does the work that wastes time. Humans make the decisions that affect safety.

01

Agents draft

A protocol goes in. Agents generate the visit-by-visit nursing sheets, eligibility checks, and safety write-ups — each citing the exact protocol section it came from.

02

QC catches, humans decide

A quality-control agent checks every draft against the source protocol and flags anything uncertain. Flagged work goes to a named human reviewer, then the coordinator and investigator. Agents never sign, dose, consent, or file on their own.

03

Signed and sealed

The investigator signs with a password and a one-time passcode. Every action is written to an append-only, tamper-evident 21 CFR Part 11 audit trail — the FDA's standard for electronic records. Nothing is deleted.

See it run

A protocol change, start to finish — with the ethics board in the loop.

Drop in a redlined amendment. TrialFlow rebuilds only the documents the change actually touches — and nothing reaches a patient until the ethics board approves it.

AI
01

Upload the change

The amendment is read and compared against the current protocol.

AI
02

Rebuild only what changed

2 of 47 documents regenerate. The other 45 stay at their approved version.

QC
03

QC flags the gap

A missing "max 800 mg" dose cap in §7.4.1 is caught against the source.

Human
04

Human clears + signs

The analyst resolves the flag; the investigator e-signs (Part 11).

Gate
05

Ethics board approves

The amended protocol and consent form go to the IRB — the required gate.

Done
06

Re-consent, safely

Once approved, affected patients re-consent at their next visit.

Nothing reaches a patient until the ethics board approves the change. TrialFlow enforces the gate — it doesn't skip it.
Both sides of the trial

The same trial, from the cancer center and from the sponsor.

Most software picks a side. A site system for the cancer center, or a separate sponsor system for the pharma company. TrialFlow runs both on one shared spine: the same trial, the same data, two views. Each person sees only what they're allowed to. Every crossover is logged.

For the cancer center — the site side
Everything the center does after a patient enrolls.
  • Generates the nursing, pharmacy, and visit documents for each trial
  • Tracks every patient's tasks, visit windows, and deadlines
  • Opens investigator-initiated trials the center's own doctors design, faster (IIT)
  • Checks eligibility and grades side effects on the standard scale (CTCAE)
  • Runs the ethics-board and re-consent cascade (IRB)
  • Files the national and cancer-center-grant reports automatically (CTRP, CCSG)
  • Reconciles billing against what each payer covers (Medicare coverage analysis)
For the sponsor — the sponsor side
Everything the sponsor runs across all its sites.
  • Sees the whole portfolio across programs, phases, and sites
  • Auto-drafts urgent safety reports on the regulatory clock (IND Safety Reports, SUSARs, DSUR)
  • Manages every IND across the portfolio: filings, amendments, annual reports, clinical-hold responses
  • Tracks ClinicalTrials.gov results deadlines before they trigger fines (FDAAA 801, up to $13,000/day)
  • Pushes one protocol change to all sites in one action (42-site network, IRB re-approval tracked per site)
  • Manages investigator brochures, 1572s, financial disclosures, delegation logs (everything BIMO inspectors ask for)
  • Logs every letter to and from FDA, EMA, and PMDA in one searchable place
  • Reconciles site budgets, milestones, invoices, and central-lab pass-through costs before paying
Trial money, reconciled on both sides. Coverage analysis on the site side. Budgets, milestones, and invoices on the sponsor side. Each side is a real reconciliation engine.
What's inside

Depth where the work actually happens.

The wedge

Document generation & QC

Visit tip sheets, nursing flow sheets, and in-service notes, written from the protocol and checked against it — line by line, with a citation for every claim.

40–80 hrs → under 2 hrs per trial (projected)
Both sides

Trial finance, site & sponsor

Site billing checked against what each payer covers; sponsor budgets, milestones, and invoices reconciled before payment.

coverage analysis · invoice-to-contract
Enroll

Eligibility

Each criterion with a plain pass / fail / needs-review and a citation.

HL7 FHIR R4 · Epic-compatible
Safety

Pharmacovigilance

Severity grading and expedited safety reports on the 7- and 15-day clocks.

CTCAE v5.0 · MedWatch 3500A
Regulatory

Submissions

Ethics-board cascade, IND lifecycle, ClinicalTrials.gov, agency correspondence.

IRB · IND · FDAAA 801
Institutional

NCI reporting

National enrollment, grant sections, and catchment & enrollment-equity.

CTRP · CCSG (P30)
Governance

Built to survive an audit, not just a demo.

The reassurance a cancer center and a sponsor both need before AI touches a trial.

E-signature, on every approval

Password, one-time passcode, and a meaning-of-signature statement — the FDA's 21 CFR Part 11 standard.

Tamper-evident audit trail

Every action writes a linked, hash-chained record. Nothing is edited or deleted; each entry traces back to its source.

A human on every output

A named reviewer signs off on every AI-drafted document. Agents never sign, dose, consent, or file on their own.

Explainable by default

Every AI answer shows its reasoning three ways — plain, clinical, and audit-ready — from one verified record.

HIPAA-architectedFHIR-native · Medplum → Epic-compatible21 CFR Part 11 audit trailNo PHI in logs or model contextSOC 2 Type II — architected
Why it's hard to copy

The moat isn't the code.

A well-funded competitor could write similar software. What they can't shortcut is everything around it.

01
One shared spine

Site and sponsor on the same system — not two products bridged after the fact.

02
Cancer-center-native depth

The NCI-specific reporting, coverage, and review work generic trial software doesn't touch.

03
Procurement + audit head start

A 9–18 month buying cycle and a 12-month security-audit clock a fast follower still has to serve.

04
Founder-market fit

Built by someone who runs these trials — operational ground truth a competitor can't hire.

Why us

Built by a physician-scientist who runs oncology trials.

MC

TrialFlow is built by Dr. Mahendra Chaudhari, a physician (MD) with a Master's in Clinical Research from Boston University. He's worked every seat in the clinical research room. Practicing physician in India for five years. Founder and operator of a clinic and a hospital. Sub-investigator on early-phase oncology trials. Today: leading and managing oncology trials at a major cancer center across NCI, SWOG, COG, Alliance, IIT, and industry-sponsored studies (sarcoma, lymphoma, cell therapy).

Every workflow in TrialFlow started as a real problem on a real trial. One he's caused, caught, or watched cause harm.

He's been building with AI in clinical research for over two years, starting with a precision-medicine program in pediatric autism. Since then: a protocol quality-control agent presented at his institution's AI Prompt-a-Thon, an oncology research-agent collaboration, 9 peer-reviewed publications, and a Cedars-Sinai Applied AI for Health Systems certificate in progress.

He's not writing about clinical research from the outside. He runs trials on Monday morning.

MDMS Clinical Research · Boston UniversityPhysician-ScientistOncology Trials LeadCedars-Sinai Applied AI · 20269 peer-reviewed publications
Get started

See TrialFlow run your protocol.

Bring a real protocol. Watch the agents draft it, a human approve it, and the whole thing get signed.