AI Data Mapping Service
The AI Data Mapping Service helps you build field-level mappings between source and target schemas faster by automatically suggesting how fields should be connected, drawing on historical mappings, semantic analysis, and generative AI capabilities.
What it does for you
When you open a mapping in Adeptia, the AI Data Mapping Service analyzes the source and target schemas you have loaded and returns a ranked set of suggestions for how fields should be mapped. Instead of manually connecting every field, you can review and apply suggestions at different confidence levels, letting the service handle the repetitive work while you focus on exceptions and business logic.
The service is useful whenever you are:
- Building a new integration that shares schema patterns with integrations you have built before.
- Mapping schemas that use different naming conventions but represent the same business data.
- Working with large schemas where manually reviewing every field pair would be time-consuming.
How suggestions are computed
The service produces suggestions through several complementary techniques, each targeting a different kind of match:
| Suggestion type | How it works |
|---|---|
| Historical (high confidence) | Finds mappings in your library that used similar source and target schemas, then checks which of their rules apply exactly to your current schemas. |
| Exact match | Identifies source and target fields whose paths match precisely, or whose leaf-level path segments match at a configurable depth. |
| Similarity match | Scores field name pairs against each other to surface fields that are likely the same concept under different names. |
| Semantic match | Uses a neural network trained on your mapping library to predict likely field pairings based on learned patterns. |
| Generative AI match | When the generative AI capability is enabled, sends field names and optional descriptions to an AI service that reasons about likely mappings and returns suggestions. If a knowledge base is connected, the service queries it to retrieve descriptions for fields, which are then included when requesting suggestions. |
| Complex match | Identifies reusable mapping rules from your library that involve multiple source fields or transformation logic, and adapts them to your current schemas. |
Each suggestion carries a confidence indicator, Confirmed for high-confidence matches and Suggested for lower-confidence ones, so you can decide how much to trust each recommendation before applying it.
How the service learns from your mappings
Every time you save a mapping and enable it for the AI library, the service indexes the schemas and field-level rules from that mapping. You can also reset the library to start fresh if needed.
What you provide
To get suggestions, you supply:
- A source schema and a target schema loaded in the mapping editor.
- Optionally, metadata tags on the mapping that help the service weight suggestions from mappings with similar business context.
- Optionally, a connected knowledge base when you want the generative AI capability to draw on your own documentation or business rules when producing suggestions.
- Optionally, field descriptions alongside field paths, which improve the quality of generative AI suggestions by giving the model richer context about what each field represents.
Applying suggestions
Once the service returns suggestions, you choose which confidence levels to apply:
- High confidence suggestions are applied as confirmed mappings.
- Low confidence suggestions are applied as suggested mappings that you can review and accept or discard.
Mapping rule suggestions
Beyond simple field-to-field connections, the service can suggest transformation rules, including XSLT-based expressions, for target fields that require logic rather than a direct copy. When this capability is enabled and a knowledge base is connected, the service retrieves transformation context from the knowledge base before generating rule suggestions, so the output reflects your organization's specific transformation patterns.
Business rules and runtime execution
The AI Data Mapping Service also supports business rule evaluation as part of your integration logic. You can define rules that validate or classify records in your data, test those rules against sample input before deploying them, and execute them at runtime as part of a running process. Rules can be organized into sets and executed in a defined order, with configurable behavior when a rule fails.
When executing a set of rules, you can control how multiple input records are merged before evaluation, choosing between an override strategy and a combine strategy, which concatenates values using a delimiter you specify.
Keeping the service in sync
The AI Data Mapping Service coordinates across all running instances of the platform. When a mapping is added to the library, when schema indexes are updated, or when the AI model is retrained, other instances are notified so that suggestion data stays consistent across the deployment.