Live demo · v0.9.2 · AI agents working in real time

Agentic Research Coordinator

Sakura's agent stack automates the data-heavy, audit-critical work of running clinical trials — autofilling CRFs, citing sources to the exact line, matching patients continuously, and keeping every site monitor-ready. Click any agent below to watch it work on real study data.

Our agent stack

Eight agents. One research workforce.

Each agent works alongside a specific role at the site — CRC, CRA, PI, Regulatory Coordinator, Scheduler, Nurse — and all eight share the same structured patient timeline, executable protocol graph, and queryable regulatory state. Click any tab to see the agent in action.

CRF Autofill Agent

Reads the patient's EHR note, reasons over the protocol's CRF schema, and populates every field with a confidence score and exact source citation — across any EDC, from any EHR.

Epic → Medidata Rave
📄 Epic · Progress Note · MRN 0049-2213-885
Dr. M. Patel · 2026-04-18 · v3 signed
🗂 CRF · Cycle 3 Day 1 · Study PEAR-204
Medidata Rave
Ready to run. Agent will extract 9 CRF fields from the progress note.

Source Citation Agent

Every CRF datapoint is linked to its exact EHR source — note ID, line number, author, signed timestamp. Part 11 compliant audit trail, always. Click any row to inspect the citation.

21 CFR Part 11 Audit trail locked
📎 CRF data points · 14 of 14 cited
100% coverage

Select a data point to inspect its source →

Every value in the CRF links back to a cryptographically-hashed EHR note excerpt, the exact line range, the signing provider, and the signed timestamp. CRAs and monitors can verify every data point without opening the EHR.

Cohort Intelligence Agent

Continuously scans every new EHR encounter against every active protocol at your site. Detects newly eligible patients, re-verification events, and exclusion shifts — before the PI's next clinic.

Live · scanning 47k patients
🎯 Active trials · live eligibility counts
Updated 4s ago

Real-time EHR event feed

east-1 · 2026-04-22

Protocol Deviation Agent

Projects every scheduled visit against its protocol-mandated window and flags predicted conflicts before they become deviations. Suggests scheduling moves the coordinator can make with one click.

3 risks · 1 conflict · next 14 days
Study PEAR-204 · visit windows · April 22 – May 5
In window
At risk
Predicted conflict

Study Startup Agent

Reads the protocol, checks feasibility against your patient population, drafts site qualification documents, and assembles a regulatory binder — reducing typical startup time from 16 weeks to under 3.

Feasibility Assessment · Protocol PEAR-204

0
High-confidence site fit.
Population match: 218 screenable patients in the last 24 months.
Infra ready. Biomarker coverage: 91%. PI bandwidth: adequate.

Regulatory binder · auto-assembly

Documents are drafted from your site's templates and populated from your CTMS, Delegation Log, and Training Matrix. PI reviews and signs — no blank-page writing.

Monitoring Readiness Agent

Continuously scores your site's audit readiness across open queries, protocol deviations, regulatory doc completeness, and source verification coverage. Builds a prioritized punch list for every upcoming monitor visit.

Sponsor CRA arrives May 8

Audit readiness index

0
of 100
Above 90 is "monitor-ready." Score updates continuously as CRFs, queries, and documents change.
📊 Breakdown · 4 categories
Computed 2m ago

🎯 Pre-monitor-visit punch list · 6 items

Scheduling Outreach Agent

The moment a patient is flagged eligible, this agent reaches out on their preferred channel — SMS, email, patient portal — answers their questions about the trial from the protocol's IRB-approved language, and books the screening visit directly into your practice management system.

Handing off from Cohort Intel
42
Eligible patients contacted this week
31
Screening visits booked (74% conversion)
2.1h
CRC scheduling time saved per day
18
Live conversations right now
📞 Outreach queue · 8 active updated 6s ago
RA
R. Almeida · MRN 0182-9901
Eligible · KEYN-B21 (TNBC adjuvant) · Preferred: SMS · English
🔒 All messages logged to CTMS · HIPAA-safe templates · IRB-approved language ehr+athena+twilio

📅 Appointments booked by Sakura today

+7 this morning

Safety Reporting Agent

The workflow that gates site capacity and trial revenue. Reads adverse event data from the EHR, codes it to MedDRA, evaluates SUSAR criteria against the Investigator's Brochure and patient history, then drafts the IND safety report — with a full justification trail for nurse, coordinator, and PI sign-off.

⚠ Adverse event detected · J. Leonard · MRN 0049-2213-885
EHR progress note (Dr. M. Patel · 2026-04-22 09:14): patient presented with fever 39.1°C, ANC 0.4 K/µL on day 7 of cycle 3. Hospitalized for IV antibiotics. Currently stable; no end-organ involvement. Receiving G-CSF per standard care.
SUSAR clock
17:42:09
until 24-hr deadline
1
Code
Reads adverse event from the EHR and maps to MedDRA terminology — System Organ Class, Preferred Term, LLT.
2
Assess
Evaluates Serious, Unexpected, and Suspected criteria against the Investigator's Brochure and patient history.
3
Report
Generates the IND safety report (CIOMS / MedWatch 3500A) with a full justification trail.
Form FDA 3500A · IND Safety Report (auto-draft)

📜 Justification trail · auto-logged

✍ Awaiting human sign-off

Sakura never files autonomously. The drafted report sits queued for the on-call nurse, the CRC, and the PI to approve. Each click hashes their e-signature into the audit trail.
Nurse
L. Carrera RN
✓ Signed · 14m ago
CRC
J. Watanabe
✓ Signed · 8m ago
Sub-I
Dr. R. Cho MD
✓ Signed · 3m ago
PI
Dr. M. Patel MD
⟳ Pending review
Shared foundation

All eight agents reason on the same substrate.

The agents aren't independent — they share a structured patient timeline, an executable protocol graph, and a queryable regulatory state. That's why a fix in one agent improves every other.

📋
Protocol Knowledge
Protocol PDFs compiled into an executable graph of visits, procedures, assessments, windows, and inclusion/exclusion rules.
🩺
Patient Timeline
Unstructured notes, labs, imaging, and meds resolved into one longitudinal, de-duplicated, clinically-typed timeline per patient.
⚖️
Regulatory Rules
GCP, Part 11, ICH, and study-specific rules encoded as constraints every agent's output must satisfy before it ships.
🧠
Clinical Reasoning
Specialist-grade clinical LLMs, grounded in the patient's timeline and the protocol graph — never free-floating inference.
Connect & structure

We meet sites where they already work.

No rip-and-replace. Sakura reads from the systems your site already uses and structures the mess so agents can act on it.

Connect

Read-only by default. FHIR, HL7, and native APIs for the largest platforms in oncology and specialty trials.

EHR
EpicOncoEMRFlatironCerner / Oracleathenahealth
EDC
Medidata RaveVeeva VaultClinical OneMedrio
Ops
Practice managementSchedulingCTMS

Structure

Raw mess in, clean structure out — the only way agents can reliably act without hallucination.

📝 Unstructured clinical notes
Longitudinal patient timeline
📄 Protocol documents (PDF)
Executable protocol graph
🗂 Regulatory binders (scans)
Queryable compliance state

Ready to see Sakura on your site's data?

We'll wire up a read-only connection to one study's EHR + EDC and run all eight agents against live data. You'll see results in under two weeks.