Build with the Infrastructure Behind Physical AI
Explore technical workflows for synthetic data, simulation, annotation, dataset quality and model evaluation.
Documentation preview: public APIs and SDKs are not yet available. Every capability below shows its real status.
- Scenario DefinitionWhat to generate or test01
- Simulation EngineVirtual environments and sensors02
- Data EngineAnnotation, QA, packaging03
- Model TrainingIn your own environment04
- Evaluation EngineMetrics, robustness, failures05
Scenario Refinement · findings feed the next scenarios
Quick start
Explore the Data Engine
Annotation formats, dataset manifests, quality assurance and versioning.
Explore Simulation Workflows
Scenario configuration, sensor concepts and sim-to-real challenges.
Explore Evaluation Workflows
Metrics by task, experiment configuration, reports and reproducibility.
Developer capabilities
- Proposed
Scenario Configuration
Describe the conditions a dataset or test must cover.
- Inputs ·
- Environment, lighting, weather, objects, sensors, seed
- Outputs ·
- Versioned scenario configuration (JSON)
- Proposed
Synthetic Data Workflows
Generate or acquire samples for an AI task under controlled variation.
- Inputs ·
- Scenario configuration, dataset requirements
- Outputs ·
- Images, sensor data and ground truth
- Proposed
Annotation and Dataset Formats
Store labels in open formats existing tooling can read.
- Inputs ·
- Samples, class definitions, conventions
- Outputs ·
- COCO-style, YOLO-style or mask annotations with metadata
- Proposed
Dataset Quality Assurance
Catch missing files, invalid classes, duplicates and mask problems.
- Inputs ·
- Dataset files and manifest
- Outputs ·
- Quality report and review list
- Proposed
Model Evaluation
Measure models overall and per scenario condition.
- Inputs ·
- Model reference, dataset version, experiment configuration
- Outputs ·
- Evaluation report with failure categories
- Proposed
Experiment Reproducibility
Make every result traceable and repeatable.
- Inputs ·
- Manifests, versions, seeds, protocols
- Outputs ·
- Experiment records that can be re-run
Architecture preview
Architecture explanation — not a live system view
Proposed designSimulation Engine
Responsibilities
- Translate scenario configurations into virtual environments
- Vary lighting, weather, objects and agents in a controlled way
- Model sensors and capture frames with ground truth
Typical inputs
- Scenario configuration (JSON)
- Sensor definitions
- Random seed
Typical outputs
- Rendered frames and sensor data
- Ground-truth labels
- Scenario metadata
Connected workflow stages
- Scenario definition (this engine)
- Simulation (this engine)
- Data preparation
- Model training (your environment)
- Evaluation
- Scenario refinement
Example workflows
Medical Imaging Research Workflow
Patient-level splits, annotation consistency checks and per-condition segmentation evaluation, as a research methodology.
Agricultural Segmentation Workflow
Scenario-driven crop and weed datasets with indexed masks, evaluated on real held-out field images.
Autonomous Perception Evaluation Workflow
Adverse-condition scenarios, sensor concepts and scenario-level detection metrics.
Documentation status
| Area | Status | Detail |
|---|---|---|
| Developer documentation (this portal) | Implemented | Guides, concepts and illustrative examples are published. |
| Illustrative examples and explorers | Implemented | Run entirely in your browser with static example data. |
| Engine workflows (Data, Simulation, Evaluation) | Proposed | Documented designs. No product implementation is available. |
| Experimental capabilities | Proposed | None are currently released for developer use. |
| Public API | Proposed | Not available. Proposed API domains are described as design categories. |
| SDKs and packages | Proposed | Not available. No package should be installed under the Synthex name. |
“Implemented” applies to the documentation and in-browser examples, not to Synthex engines. See the API reference and SDKs pages.
Technical pilots
Explore a Technical Pilot with Synthex
Working on a dataset, simulation or model evaluation problem? Describe it, and we will discuss whether a focused technical pilot could help. No public API or SDK is needed to start.
Prefer the product view? See the platform overview.