Medallion architecture

RebelCore organises tabular data work into four stages. Source documents use the same governed module and project boundary but do not need a tabular Tree.

Add source data → Create a module → Curate tabular data → Ask Agent

The terms Bronze, Silver, and Gold may appear in the product and documentation:

Stage	RebelCore action	Outcome
Bronze	Add Data	Tabular files and source documents are received and classified.
Silver	Create a module	Tabular data is organised; documents are parsed and vectorised.
Gold	Curate the Tree and build a semantic dataset	A focused tabular or mixed selection is ready for Agent. File-only modules skip this stage.
Agent	Start a session and ask questions	Agent answers within the selected access, exposure, and privacy controls.

1. Add source data

Open Datasets, select Add Data, name the module, and add Excel, CSV, or supported source documents. RebelCore separates tabular data from source files and asks you to resolve ambiguous items.

Choose the LLM exposure that should follow the resulting module.

Add files

2. Create a module

When tabular data is present, open Module Builder. Review the schema, add the required labels, confirm the exposure level, and build.

For documents only, RebelCore creates the file module and schedules vectorisation without opening Module Builder. Both paths continue into project creation.

Create your dataset

3. Curate and build

Create or open a project with tabular capability, then use the Tree to:

• Navigate the dataset structure.
• Select the fields relevant to the intended questions.
• Review available data suggestions.
• Build one or more semantic datasets.

How the Tree works

File-only projects open directly in Agent because there are no columns or weights to train.

4. Ask Agent

Open RebelCore Agent, select the datasets required for the task, choose a response mode, set the privacy controls, and submit your question.

For mixed datasets, one selection can provide both structured-data and source-document context. The most restrictive applicable exposure governs the response.

RebelCore Agent

Access at each stage

Roles can separate these responsibilities:

• A data steward can add data and build modules.
• A project analyst can curate projects and semantic datasets.
• An Agent user can ask questions without receiving data-administration access.
• The Customer Super Admin can inspect personal shares and dataset-derived file access.

See Governance & access and Roles & permissions.

Fix problems at the right stage

• A file or column is missing: review the Datasets module detail and its separate tabular/source-file groups.
• The labels or module are wrong: rebuild the module.
• The Tree includes irrelevant fields: adjust the Tree selection.
• Agent cannot find a document: confirm vectorisation completed, the correct dataset is selected, and file content is enabled.
• Agent cannot find tabular information: confirm the semantic dataset is ready and selected.
• Agent returns less detail than expected: check the effective LLM exposure level.