A clinical study, from authoring to SDTM, on your laptop.
The Provenance Neural Engine™ Community Edition is a free, open reference deployment of a domain-agnostic data engine on an NGSI-LD broker, demonstrated end to end on clinical research, with a sprinkle of AI. No cloud dependency of any kind: no cloud account, no message queue, no object store to provision.
Requires Docker. Apache-2.0. The AI step runs offline by default; add an Anthropic API key when you want a real model.
npx pne-community-edition or a plain git clone.What happens when you run it
Six steps, in order, each one a real request against a real service. The walkthrough UI drives them; every step is also plain curl.
Author
A USDM-style study definition becomes NGSI-LD entities in an upstream NEC Scorpio broker, typed with the open T.O.P. cr-domain vocabulary: Study, Arms, Endpoints, EligibilityCriteria, Visits.
Project
The study design is projected to CDISC SDTM trial-design domains (TS, TA, TI, TV) as Dataset-JSON. Standards are views, not the stored model.
React
Every write to the broker fires an NGSI-LD subscription that notifies the pipeline directly over HTTP. That one hop replaces SNS, SQS, and a Lambda.
Transform
sdtm.oak, the open-source CDISC/pharmaverse engine, maps raw execution data to SDTM: controlled terminology, sequence derivation, the real thing.
Query
DuckDB runs SQL straight over the files in the lake. Athena semantics, zero infrastructure. You can also just
lsthe lake.Sprinkle of AI
A natural-language clinical statement becomes a broker entity and re-triggers the whole pipeline. One model call, or a deterministic offline stub when there is no API key.
Cloud service, replaced
The architecture (an NGSI-LD broker fronting a CDISC pipeline) is inspired by the AWS Garnet Framework, and built without it. The broker itself never needed the cloud: NEC Scorpio ships an in-memory build that runs every broker service in one JVM. SQS was a packaging choice, not a requirement.
| Garnet used (AWS) | Community Edition uses |
|---|---|
| SQS/SNS subscription fan-out | NGSI-LD subscriptions delivered straight to a local HTTP listener |
| S3 object persistence | A filesystem directory you can ls |
| Athena SQL over the lake | DuckDB over the files |
| Kinesis Firehose | Direct writes from the pipeline |
| Fargate orchestration | docker compose |
Need parts of the cloud back for a production deployment? The lake is deliberately swappable: Postgres, Athena, Trino, ClickHouse, MinIO all fit behind the same small contract.
One engine, many domains
Clinical research is the first anchored domain, not the boundary. Domains are pluggable vocabularies from The Ontology Project; the broker, context, subscription, projection, and lake architecture stays the same. The applicability runs across the healthcare and life sciences value chain, from healthcare delivery through drug development, commercialization, and claims, and into the territory that CMC naturally crosses: manufacturing, quality, and supply chain. Swap the vocabulary and the projections; keep the engine.
What this edition covers
The Community Edition is a complete local reference stack: an off-the-shelf NGSI-LD broker, an open transform engine, Dataset-JSON and SDTM projections, and one optional model call (or an offline stub). It is built for local practice and learning on a single node.
| Capability | Community Edition |
|---|---|
| License | Apache-2.0 |
| NGSI-LD broker + CDISC pipeline (USDM, Dataset-JSON, SDTM) | Yes |
| AI entity construction | One model call or offline stub |
| Provenance and time | Broker-native temporal API |
| Deployment | Laptop, single node via docker compose |
Learn more about Scientix.AI at scientix.ai.
Built in the open, on open things
Nothing in this stack is proprietary. Architectural inspiration is credited to the AWS Garnet Framework; none of its code or services are used.