Add documentation for MLOps Workflow features and best practices

SUMMARY.md

@@ -77,6 +77,7 @@
   * [Collections](data-organization/project-dataset/collections.md)
   * [Quality Assurance & Statistics](data-organization/project-dataset/quality-assurance-and-statistics/README.md)
     * [Practical applications of statistics](data-organization/project-dataset/quality-assurance-and-statistics/practical-applications-of-statistics.md)
+  * [Data Versions](data-organization/project-dataset/data-versions-mlops.md)
 * [MLOps Workflow](data-organization/mlops-workflow.md)
 * [Team Files](data-organization/team-files/README.md)
 * [Disk usage & Cleanup](data-organization/storage/README.md)

data-organization/project-dataset/data-versions-mlops.md

@@ -0,0 +1,99 @@
+---
+description: >-
+  The main features, capabilities and usage recommendations of MLOps Workflow are described in this documentation.
+---
+
+# What is MLOps Workflow in Supervisely
+
+**MLOps Workflow** in Supervisely helps you:
+
+1. **Track data changes** – create versions of a project and monitor what changed and when.
+2. **Ensure experiment reproducibility** – return to an exact state of data and models.
+3. **Collaborate as a team** – see how the project evolved and who ran what.
+4. **Understand the process easily** – visualize the entire data/model pipeline.
+
+## 1. Data Version Control
+
+- You can save any project state (after uploading, annotating, filtering, etc.) as a **version**.
+- Versions are created **automatically** when you run apps like TrainYOLO or other training-related apps.
+- This ensures you can **reproduce an experiment** later using the same data state.
+- You can also **manually create a version**, but **only if there were changes** in the data. Otherwise, it won’t make sense and the system will prevent it.
+- The **only exception** is the first version — you can always create it because there’s no prior state to compare to.
+- While versions **can be deleted**, it is **not recommended** — the project history might be lost.
+- You may notice **intermediate versions** in the MLOps Workflow graph, but they are just **visual indicators** and do not appear in the project’s version tab.
+
+## 2. Visualizing the Full Process
+
+- The interface shows a **graph** – with blocks representing data and applications.
+- A block may represent data (projects, datasets, files) **before or after processing**, or an operation/app.
+- Example: upload images → save version → train YOLO → get model and report → create a new version.
+- Each step is clickable — you can view, download, or recover results at any point.
+
+## 3. Model Evaluation (Benchmarking)
+
+- After training, you automatically get a report with:
+  - mAP, Precision, Recall, IoU
+  - Inference speed, and more
+- These metrics help you compare different models or data setups.
+- Easy side-by-side comparison of model versions.
+
+## 4. Workflow Building – Best Practices
+
+- Prefer working with **projects**, not just datasets — the graph will be easier to follow.
+- **Group related steps** (e.g., apps in folders).
+- Add **descriptions** to blocks — teammates will understand your logic faster.
+- Avoid **overwriting existing versions** — keep the process traceable.
+- When launching apps, **preserve session context**, don’t run everything in one long chain.
+
+## 5. Getting Started
+
+You can open the MLOps Workflow interface from:
+- The **project menu**
+- **Tasks** section
+- Directly from the **workspace**
+
+## 6. Integrating Workflow into Custom Applications
+
+If you're building your own app using the Python SDK:
+
+- Initialize with:
+  ```python
+  api = Api.from_env()
+
+This will enable api.app.workflow in your app context.
+
+Then use methods like:
+```
+  api.app.workflow.add_input_project(project, version_id=...)
+api.app.workflow.add_output_project(...)
+```
+
+These calls add blocks to the workflow graph showing your app’s input/output.
+
+There are also methods for datasets, files, sessions, labeling jobs, etc.
+
+It’s a simple way to visualize your app’s logic in the MLOps Workflow.
+
+## 7. Why Data Versioning Matters
+
+To summarize:
+
+Automatic versions are created by training-related apps (e.g., TrainYOLO).
+
+- Manual versions require data changes to be saved.
+
+- Version 1 can always be created — it’s your baseline.
+
+- Deleting versions is possible but not encouraged.
+
+- Intermediate versions may appear in the graph view but not in the project’s version list — they are for orientation only.
+
+## Why MLOps Workflow Is Important
+
+- Enables you to revert to any data state when needed.
+
+- Helps analyze performance — what change made the model better?
+
+- Simplifies teamwork — every step is visible.
+
+- Speeds up development and delivery — everything is traceable and repeatable.