DataFact workflow overview

DataFact organizes work into five connected stages: connect source material, turn it into reusable MemoryLake resources, add business knowledge, use those resources in AI applications, and evaluate the results.

You do not need every stage for every task. Start with the shortest path that matches your source and desired output.

flowchart TD
    A["Connect documents or databases"] --> B["Build MemoryLake resources: Datasets, Documents (Lite), or Documents (Excel)"]
    B --> C["Model knowledge with dataset knowledge or Analysis Models"]
    C --> D["Build AI applications with Agents, AI Reports, or Channels"]
    D --> E["Evaluate and improve with Feedback, Q&A Management, or Test Sets"]
    E -. "Improve knowledge and configuration" .-> C

Availability

The entries you see depend on the capabilities available in your workspace and on your permission to use them. If an entry mentioned here is not visible, use another available workflow or ask the person responsible for your DataFact workspace which path you should use.

1. Connect source material

Connections bring source material into DataFact without yet defining how an agent should use it.

• Use Document Library to upload and organize files that can later be added to document memory.
• Use Databases to create and test a database connection, then inspect its schemas and tables.

At the end of this stage, DataFact can reach the source. The source is not automatically ready for every analysis or question-answering workflow.

2. Build MemoryLake

MemoryLake turns connected material into resources that AI applications can use.

Choose a resource by the structure of the source and the task you want to complete—not only by the original file extension.

3. Model knowledge

Knowledge helps DataFact interpret source material using the language and rules of your business.

For a Dataset, you can maintain knowledge such as business entities, relationships, business knowledge, general knowledge, and optimization examples. When a compatible Analysis Model is bound to a Dataset, the model becomes the analysis path for that Dataset; the Dataset's own knowledge remains visible but may become read-only.

Analysis Models add versioned knowledge building. A completed version can be selected for use, while versions that are still building or have failed are not editable as completed versions.

4. Build AI applications

AI applications give people a way to use the prepared data and knowledge.

• Agents provide a conversational workspace for questions and results.
• AI Reports use templates and selected spreadsheet content to request a structured report where the capability is available.
• Channels publish an agent to a supported messaging destination where the capability is available.

An application does not make an unavailable source accessible. Its answers and outputs depend on the resources connected to it and on the access available to the current user.

5. Evaluate and improve

Evaluation turns real usage into evidence for the next improvement.

• Feedback records a user's assessment of an answer when feedback controls are available.
• Q&A Management helps review visible questions and feedback records.
• Test Sets run reusable questions so results can be compared after changes.

Evaluation does not automatically guarantee a better answer. Use the findings to revise knowledge or application configuration, then run the relevant questions again.

Choose a starting path

Related guides

• Get your first answer in 5 minutes
• The database, document, knowledge modeling, and evaluation tutorials remain in the editorial backlog until their workflows pass the same evidence review.