An introduction to Astra

Clinical trial reporting, under a clearer lens.

Astra brings six AI specialists to public trial records and research abstracts. It looks for gaps between what was registered, what was reported, and what was published—then makes the evidence available for human review.

Explore the approach Meet the specialists

Built around public ClinicalTrials.gov records and PubMed abstracts

Astra / analysis path01—04

A question becomes an evidence-led brief.

Public sources

ClinicalTrials.govPubMed

Specialists investigateRelevant agents work in parallel

Evidence is checkedClaims are validated against records

Human judgment stays in the loopUncertain findings wait for review

Signals are leads for human review, not findings of misconduct.

Astra’s guiding principle

01 / The approach

Facts first.
Judgment second.

Astra is designed to make an investigation understandable, from the original public record to the reason a signal was raised.

Compute

Python and SQL calculate dates, reporting rates, outcome comparisons, and adverse event counts.

Interpret

Specialist agents assess those facts, explain what may matter, and cite the relevant trial or paper.

Check

A validator screens unsupported signals before results are saved or sent for review.

02 / The specialists

Six ways to look closer.

Each specialist asks a different question about the same public evidence. The supervisor selects only the ones relevant to a task.

Missing Results

Looks for completed trials whose required results are overdue.

Broken Promises

Compares registered primary outcomes with results and linked papers.

Track Record

Examines how reliably a sponsor posts results when they are due.

Pattern Finder

Finds reporting patterns that stand apart from comparable studies.

Side Effect Checker

Checks whether serious adverse events appear in the registry but not linked abstracts.

Timeline Analyst

Spots studies that have drifted well past their own stated schedule.

03 / Human review

A signal is a starting point, not a verdict.

Findings carry citations and confidence scores. Lower-confidence signals enter a review queue; when a reviewer rejects one, Astra turns that feedback into a rule for future runs.

Designed for scrutiny.

Evidence can be checked. Decisions can be revisited. The system learns from corrections.

What comes next

The full workspace
is on its way.

This page introduces the research behind Astra. The interactive workspace for exploring signals, watching agents run, and reviewing evidence will follow.

Explore the live API