progenitor
Research demoSynthetic patients only

Your next enrolment starts here

Read the record.Find the right trial.Explain why.

A team of AI agents reads a patient's health record, screens live recruiting trials, and checks every eligibility rule against cited evidence. A clinician approves the shortlist.

  • Every verdict cites the record
  • Clinician approves every shortlist

A long record. A short, explained list.

progenitorJordan Reyes, 58Synthetic recordInteractive preview
patient.json01 / Read the record
  1. {
  2. "resourceType": "Patient",
  3. "id": "jordan-reyes",
  4. "gender": "male",
  5. "conditions": [
  6. { "code": "SNOMED 44054006", "text": "Type 2 diabetes mellitus", "onset": "2019-04" },
  7. { "code": "SNOMED 38341003", "text": "Hypertension", "onset": "2021-08" }
  8. ],
  9. "medications": [
  10. { "rxnorm": "6809", "text": "metformin 1000 mg twice daily", "since": "2024-09" },
  11. { "rxnorm": "29046", "text": "lisinopril 10 mg daily", "since": "2021-09" }
  12. ],
  13. "labs": [
  14. { "loinc": "4548-4", "text": "HbA1c", "value": "8.4 %", "date": "2026-03-02" },
  15. { "loinc": "98979-8", "text": "eGFR", "value": "72 mL/min/1.73m2", "date": "2026-03-02" },
  16. { "loinc": "39156-5", "text": "BMI", "value": "31.2 kg/m2", "date": "2026-02-11" }
  17. ]
  18. }
Explained shortlist
  1. NCT0XXXX101 · Recruiting

    Once-weekly GLP-1 therapy added to metformin in type 2 diabetes

    • Met:Age 18 to 75Patient/jordan-reyes
    • Met:HbA1c 7.5 to 10.5 %Observation/hba1c-2026-03
    • Met:Stable metformin for 3 months or moreMedicationRequest/metformin-2024-09
    • Met:Exclusion: insulin in the last 3 monthsMedicationRequest?code=insulin
  2. NCT0XXXX102 · Recruiting

    Coached continuous glucose monitoring for adults with obesity and diabetes

    • Met:BMI 30 or moreObservation/bmi-2026-02
    • Unknown:Owns a smartphoneno evidence in record
  3. NCT0XXXX103 · Recruiting

    SGLT2 inhibitor in diabetic kidney disease

    • Not met:eGFR 25 to 60Observation/egfr-2026-03
    • Unknown:Urine albumin-to-creatinine ratio 300 mg/g or moreno evidence in record
02 / Your explained shortlist

An illustrative preview with synthetic data. Research demo. Not for clinical use. Open Progenitor to run a real match on a synthetic patient. Try another patient above.

Built on open standards

  • FHIR R4
  • ClinicalTrials.gov
  • PubMed
  • RxNorm
  • MCP
  • Bedrock AgentCore

Less reading. More enrolling.

From a long recordto a short list.

Each agent does one job. An AI supervisor decides what happens next.

  • Patient Profiler

    Reads conditions, medications, and labs from a FHIR server, with codes and dates.

  • Trial Scout

    Searches live recruiting trials on ClinicalTrials.gov and refines the search when results are thin.

  • Criteria Parser

    Turns free-text eligibility rules into checks a program can test.

  • Eligibility Judge

    Marks each rule met, not met, or unknown, and cites the record for every verdict.

  • Evidence Agent

    Adds supporting papers from PubMed to each shortlisted trial.

  • AI Supervisor

    Decides what to fetch, which trials to pursue, and when to ask a clinician. Every decision is logged.

Clinicians stay in charge

Same patient.Same page.

Coordinators start a match. Clinicians answer open questions, review each verdict with its evidence, and approve the shortlist before any referral is drafted.

See the review flow

Sign in to review and approve.

review / NCT0XXXXXXXAll decisions logged

Exclusion: insulin in the last 3 months

The Judge found no MedicationRequest for insulin since 2025-11Judge

Dr. Rao · clinician

Confirmed with the patient. Approve.

Reply in context

An example of a clinician review

Every decision on the record.

The supervisor's choices, with reasons, are saved as they happen. Replay a run step by step.

  1. Record fetched
  2. Search broadened
  3. Shortlist ranked

Built for the referral

Ready for“refer patient.”

Explained shortlists, linked evidence, and a referral draft waiting for sign-off.

See a sample shortlist
  • Explained shortlists

    Each trial lists every rule with met, not met, or unknown.

  • Evidence you can open

    Every verdict links to the FHIR resource it came from.

  • Referral drafts

    Drafted only after a clinician approves. Never sent automatically.

  • Measured, not guessed

    Evaluated against published NIH TrialGPT labels.

Runs where your data lives

Your agents.Your cloud.Your rules.

Run locally with Docker on synthetic data, or deploy the agent graph to Amazon Bedrock AgentCore inside your own AWS account.

Self-hosted tracing. No patient data leaves your environment.

Read the deployment guide
progenitorrun / jordan-reyesApproved
  1. fetch_recordPatient/jordan-reyes
  2. search_trialsT2DM, recruiting
  3. broaden_search+GLP-1, 50 mi
  4. judge_trials3 trials, 14 rules
  5. rank_shortlistby met rules
  6. ask_clinicianinsulin history

Shortlist

  • NCT0XXXX101
  • NCT0XXXX102
  • NCT0XXXX103
  • Docker
  • AgentCore
  • Local
Same graph, local or in the cloud.

For the way you build

Bring your ownagent.

Progenitor's trial search and terminology tools are MCP servers. Connect Claude Code, Cursor, or another MCP client and use them directly.

  • Read-only tools by default.

Explore the MCP tools

Terminal
$ uv run python -m mcp_servers.clinicaltrials

A long record. A short list.

Every enrolmentstarts with a match.

Open Progenitor. Pick a synthetic patient. See which trials fit, and why.

Match a patient

Synthetic data only. No account needed to view the demo.