In Vivo

Live registry prototype

Clinical Pipeline & Recruitment Atlas

Methodology

Filters

Loading metadata...

Pipeline summary

Therapeutic areas

Trials or candidates can appear in more than one area.

Area Candidates Share Trials Median trial age Recruiting 24+ mo.
Recruitment explorer

Result set

Candidates

Candidate Lead sponsor Mechanism / class Trials Median trial age
Extraction quality

Eligibility extraction eval

Field Correct Errors Accuracy

Material findings

Definitions

Metrics

Backend runs

ClinicalTrials.gov ingestion

NCT ID Title Status Sponsor

LangGraph + OpenRouter

Build walkthrough live census + reviewed AI sample
  1. Front end

    This page is the viewer. It reads JSON files and turns them into filters, cards, tables, and drill-downs.

  2. Raw data

    Python fetched every registry record in the current scope from ClinicalTrials.gov. Historical snapshots remain a separate extension.

  3. Cleaning

    Python standardizes phases, sponsors, conditions, interventions, dates, and current trial statuses.

  4. Agent loop

    LangGraph extracts structured eligibility fields, validates them against source text, runs an independent model judge, and routes each record to approval or review.

  5. Evals

    Two independent reviews were reconciled into pilot gold labels. Ambiguous source records are not scored.

  6. Export

    The builder writes summary, area, candidate, trial, definition, metadata, and evaluation JSON files.

  7. Publish

    The static files are deployed to the VPS. Readers see the dashboard; they do not trigger live AI calls.

Context static viewer + offline builder