AI & Machine Learning · Enterprise Software · Healthcare

nPhase.ai — Clinical Data Platform & AI Statistical Computing

From a high-volume clinical data platform to the AI that turns raw trial data into submission-ready datasets and statistical programs — a long-running engagement with nPhase.ai, formerly RedCap Cloud.

Both ends of the clinical data problem

nPhase.ai, formerly RedCap Cloud, has had us working at both ends of the life of clinical trial data. At one end, large volumes of research data arrive from many sources and have to be transformed and made available for analytics. Traditional relational databases couldn’t handle the analytical query workloads, and every new data source meant another round of pipeline engineering.

At the other end is the work that turns collected trial data into an FDA submission: mapping it to CDISC standards, deriving analysis datasets, programming every table, listing and figure, and validating the result. It is specialist, slow and largely manual, and every step has to stand up to 21 CFR Part 11 scrutiny.

The data platform: pipelines in hours, not weeks

We built a visual, drag-and-drop ETL tool for building pipelines without code, on ClickHouse as the OLAP warehouse, Kafka for real-time streaming and Kubernetes to orchestrate the microservices. Data comes in through connectors for clinical research systems, including REDCap (Research Electronic Data Capture), runs through configurable pipelines and is served through analytics dashboards.

  • A new pipeline takes hours to build instead of weeks
  • Analytical queries that used to take minutes return in under a second on ClickHouse
  • The Kubernetes deployment scales as data volumes grow

The Statistical Computing Environment: AI inside a governed workflow

For nPhase’s Statistical Computing Environment (SCE), we build the platform that carries clinical data from raw collection to submission-ready output. It is multi-tenant and built for 21 CFR Part 11, and AI sits inside the workflow rather than beside it:

  • SDTM mapping, ADaM dataset generation and table, listing and figure (TLF) creation, with integrated Pinnacle 21 validation
  • AI-generated statistical programs for TLFs and ADaM datasets
  • A vision model that extracts table layouts from shell documents into an editable design
  • Conversion of the statistical analysis plan to the Analysis Results Standard (ARS)

Every output has Pinnacle 21 validation and a full audit trail behind it.

An assistant that asks before it acts

An AI assistant works alongside users inside the application. It answers questions about projects and studies through a controlled read layer rather than raw database access, and it can drive the interface — navigating, opening dialogs, filling in fields. Anything that changes durable state, such as saving, submitting or deleting, needs the user’s explicit confirmation, and attachments shared with it are treated as untrusted input. In a regulated environment, that guardrail is the difference between an assistant and a liability.

Building something similar?

If you are planning a clinical data platform or AI for clinical trials, book a 30-minute call. We will walk you through how we built this one, what it would take for yours, and who would work on it.

Book a 30-minute call More case studies