Cloud-Native Workflow Platform

Context

Most of the current commercial workflow-oriented solutions for geo-model building consist of sets of monolithic applications. As they are, they cannot interact with external products like proprietary applications or innovative applications proposed by SMEs or research institutes.

A solution for overcoming this difficulty is to consider that workflows should no longer be monolithic but should consist of a set of interoperable atomic activities (micro-services). Modelers should instead be offered the possibility of accessing a Cloud Workflow Platform on which they could freely identify, select and assemble cloud-nested applications in order to fit the various geo-modeling tasks they intend to perform.

A geological model is never built in one pass. A geologist looks at an intermediate result, decides, and the work carries on — or goes back one step. And when new data arrive, the model must be updated rather than rebuilt. A workflow engine whose parameters can only be set at launch cannot follow this way of working.

The platform

Geosiris develops an open-source based Cloud-native workflow platform that orchestrates geomodeling business activities described in the BPMN 2.0 standard. A workflow combines two kinds of tasks: service tasks, executed automatically by geomodeling services, and user tasks, where the geologist checks a result and takes a decision.

• Decisions in the loop. A workflow can be interrupted, resumed and steered at any moment while it runs, not only parameterised at launch.
• Cyclic workflows. A result judged unsatisfactory sends the workflow back to the step that produced it, with new parameters, until it is accepted.
• Update, do not rebuild. When new data arrive, only the activities concerned are run again and the model is updated in place — the principle of a geomodeling digital twin.
• Everything is traced. Each run is recorded as an Energistics/OSDU Activity, with its inputs, outputs and parameters. The history tells how every object of the model was obtained, and allows an inconsistency to be found and corrected by simply relaunching the workflow with new inputs.

End users can reuse workflows provided by Geosiris or its partners (“menu solution”), or assemble their own from the catalogue of services (“à la carte solution”).

The platform has been evaluated in four domains: oil and gas, geothermal energy, geotechnics and geological hazard assessment.

Architecture of the Geosiris solution

1. A BPMN workflow orchestrator

• Developed by Geosiris on the open-source SpiffWorkflow engine, it runs the BPMN diagrams, distributes the service tasks and presents the user tasks to the right people.
• A PostgreSQL database holds the definition of the services, the roles of the users and the follow-up of every task.

2. A catalogue of geomodeling services

• Each service is described as an Energistics/OSDU Activity Template: its name, its semantic description, its keywords, and its inputs and outputs — RESQML/OSDU objects or parameters.
• The services run in Docker containers inside a Kubernetes cluster, isolated from the outside world by a gateway. They are triggered by events, and their resources scale automatically with the load. Existing code in Rust, Python, Java or C++ can be reused as it is.
• Services provided by AI agents or hosted on external scientific workflow platforms such as n8n or Galaxy can also be called, through a gateway based on OGC standards, without the user having to know how they are implemented.

3. Standard-based data

• The data are held as RESQML dataspaces as a series of EPC + HDF5 files or on an ETP server — GABBRO, our implementation, equivalent to the OSDU Reservoir DDMS — and exchanged in real time over the ETP protocol (WebSockets).
• Within an OSDU environment, the data catalogue is queried first, so that only certified (“trusted”) entities enter a workflow; enriched results are written back to the catalogue and can in turn be certified.

4. A memory of experience

• Every operation is recorded and indexed, so that past runs can be searched and reused. Access is managed per user.

5. Two applications

• A workflow modeler, our add-on to BPMN.io, for the experts who design the workflows: a no-code tool to draw the BPMN diagrams, assemble the services and build the forms of the user tasks.
• GEOSIRIS-FLOW, the web cockpit of the business user, to launch the workflows, follow them and answer their decisions.

What has been demonstrated

• Building an earth model from borehole data. A first workflow chains three business sub-processes: transcription of borehole data into RESQML objects, exploitation of the well markers, and construction of the contact surface between geological units by kriging — with a quality control and a decision of the geologist at each step. Video
• Continuous update of the model. When the surface is judged unsatisfactory, the computation is rejected and relaunched automatically with new parameters; when new boreholes arrive, the model is updated rather than rebuilt, and exported to third-party tools (GIS, BIM, 3D viewers). Video
• Interactive fault modeling on OSDU data. A ten-task workflow takes certified data from the OSDU catalogue, lets the geologist choose the faults to model, computes them by multilevel B-spline approximation, checks and displays the result, and writes the enriched entities back to the catalogue for certification — an OSDU partition maintained as the digital twin of a prospect. Video · EAGE 2026 talk

These results have been presented at the EAGE Annual Conferences (2024, 2025 and 2026) and at the Journées Nationales de Géologie et Géotechnique (2026): see Publications / References. Recordings: EAGE 2024 demonstration, EAGE 2026 talk.

Who it is for

The platform addresses in priority operators’ corporate research departments, geological surveys, engineering companies, small ISVs (Independent Software Vendors) and universities. We offer these partners the possibility of co-developing it with us, and of using it to distribute their own geomodeling services in a seamless, standard-based environment — and to build numerical shadows of earth models that can become digital twins by adding AI agent operations (n8n agents running sub-workflows).