Anurag Jha
← All products

NeuroDiscovery AI · 2026 · B2B · Clinical trial recruitment

Clinical Trials Portal

internally, Patient Reviewer

AI identified trial candidates go to research sites; site staff move each patient from eligible to enrolled, with every change audited and identity shown only to the right roles.

My roleProduct owner: roadmap, specs, new features in the living prototype, the stability release and incident analysis.

3research sites live since June 2026Product roadmap
~6 wksfrom build start to live, 40+ ticketsBuild epic
1 dayto close a partner trust incident, using the audit logIncident analysis
One trial, eligible to enrolled. Demo data.

The challenge

Recruitment ran on spreadsheets between our team and research sites, with no audit trail and no safe way to share who a patient is.

What shipped

Candidate upload with validation, a five stage pipeline, feedback by role, a survey project type, and five roles across two organisations enforced on the server.

What moved

Live at 3 research sites within about 6 weeks. Patient lists that took 5 to 10 minutes to load were fixed, and a partner trust incident was closed in a day

A

From list to enrolment

Upload, validate, move each patient through five stages, and keep a trail.

  1. 01

    Sign in and see your trials

    Single sign on; site staff only see the trials assigned to them.

    Sign in.
    All trials. Demo data.
    01 / 02
  2. 02

    Start a project

    A clinical trial or, since July, a survey project.

    Project type.
  3. 03

    Upload and validate

    Lists are parsed, validated and deduplicated. A bad file is rejected with an error report before anything is saved.

    Upload.
    Validation.
    01 / 02
  4. 04

    Move patients through the funnel

    Eligibility, outreach, consult, screening, enrolment. Each stage has defined states.

    The pipeline.
    Filter by stage.
    Update a stage.
    01 / 03
  5. 05

    Feedback with a trail

    Every note is tied to a role and a time stamp.

    Feedback trail.
  6. 06

    Surveys

    Screening, reminders, completion and incentives as a second project type.

    Survey setup.
    Survey review.
    01 / 02

How it was evaluated

  • User acceptance checklists across nine record tabs
  • Access level QA with test accounts before every list is shared
  • Post release checks with the site on load time, counts and comments

Candidates are identified upstream, so no AI accuracy is claimed here. Enrolment outcomes are not available to me.

Technical choices

  • React and TypeScript, NestJS services, PostgreSQL with an encrypted identity schema, OIDC single sign on, Kubernetes on AWS

What I learned

  • In regulated work the audit log is a product feature.
  • Privacy by role must be enforced on the server.