Astraea logoAstraea
Book a demo

Our story

Astraea began with a recurring question from people close to us in pharma: why did clinical analysis still depend on so many manual steps and disconnected handoffs?

Joshua Wang and Sanmay Sarada brought together experience in production AI and clinical data infrastructure to build a connected biometrics workflow. The company started with oncology and rare disease, where complex analyses make every handoff consequential.

Today, Astraea brings standards, statistical programming, quality control, and review into software that sponsor teams operate in their own environment. Experts retain the decisions; automation supports the work around them.

Our work covers the steps after a protocol is established, through submission preparation. Astraea supports statistical analysis plans, electronic case report forms, standardized study datasets (SDTM), analysis datasets (ADaM), tables, listings, and figures. The scope also includes Define-XML metadata, quality control, and submission documents. These deliverables depend on one another, so our approach connects their execution and review rather than treating each as an isolated task.

Astraea is a software platform, not a contract research organization. Forward-deployed engineers install it inside the customer’s own environment, rather than asking study teams to upload clinical data to an Astraea-hosted service. Astraea does not see customer data. Sponsors retain control of their infrastructure and trial responsibilities, while qualified experts remain accountable for scientific interpretation and acceptance of the outputs.

Our story on Y Combinator

How we build

Driven by intelligence. Grounded in responsibility.

  1. 01

    Design for the reviewer

    Make the evidence, assumptions, and next decision easy to understand. A reviewer needs the source data and analysis logic behind an output, not only a finished table or figure. That context supports an informed acceptance decision.

  2. 02

    Keep the science in view

    Automate repetitive work while preserving expert judgment. Study-specific endpoints, population definitions, and statistical methods still need qualified interpretation. Automation supports execution and checking; it does not replace the people responsible for the science.

  3. 03

    Build for continuity

    Connect decisions across documents, datasets, and outputs. A change to an analysis rule can affect several deliverables. Keeping those dependencies in view helps teams assess what needs to be updated and reviewed together.

  4. 04

    Respect the boundary

    Bring software to the client’s environment and governance. Deployment planning should establish access permissions, validation scope, and operational responsibilities with the sponsor’s IT, security, and quality teams before clinical data enters the workflow.

  5. 05

    Make progress visible

    Show what changed, what was checked, and what needs review. A detected finding is not the same as a resolved issue. Keep corrections and reviewer decisions connected so teams can distinguish completed work from outstanding questions.