AI-Assisted Data Mapping
The Adeptia platform includes an AI-powered data mapping capability that automatically suggests how fields from a source schema should be mapped to fields in a target schema, reducing the manual effort required to build and maintain integrations.
What it does
When you provide source and target schemas, the AI agent analyzes both layouts and produces field-level mapping suggestions, including the transformation logic needed to reconcile differences in structure, naming, or data type. The agent works through the mapping in stages, emitting progress updates as it goes, so you can follow along in real time rather than waiting for a batch result.
The capability is designed to fit naturally into the integration-building workflow: you supply the schemas you are working with, optionally provide additional context or instructions, and the agent returns a set of mapping recommendations you can review and refine.
What you provide
When starting a mapping task, you identify the source and target schemas you want to map between. You can also supply:
| Input | Purpose |
|---|---|
| Source schema identifiers | The schemas the agent reads data from |
| Target schema identifiers | The schemas the agent maps data into |
| Specific unmapped target fields | Focus the agent on fields that still need a mapping |
| A selected target field | Direct the agent to concentrate on one particular field |
| Additional instructions | Plain-language guidance to shape how the agent approaches the mapping |
| Context files | Supporting documents such as field definitions, business rules, sample data, or mapping specifications |
Context files can be uploaded ahead of time so they are ready when the mapping task starts. Supported file types include CSV, JSON, XML, Excel spreadsheets, PDF documents, and Word documents.
Real-time progress and streaming
The mapping agent streams progress events as it works, so the product UI can display live status updates. Events cover each stage of the process, including layout analysis, field-by-field mapping decisions, transformation generation, and completion. If the connection is interrupted, the stream supports reconnection using the last received event identifier.
Human review and intervention
The agent is designed to pause and ask for your input when it encounters a mapping it cannot resolve with sufficient confidence. These pauses are called interventions. At each intervention point, the agent presents the specific fields or structural gaps it needs help with, along with suggestions and possible source fields to choose from.
For each gap, you can:
| Action | What it means |
|---|---|
| Select a source field | Choose which source field should map to the target |
| Provide a default value | Supply a fixed value when no source field applies |
| Provide transformation logic | Describe how the source value should be converted |
| Add a note | Record your reasoning for future reference |
You can also provide general feedback that applies across all open gaps at once, giving the agent broader guidance before it continues.
Once you submit your decisions, the agent resumes processing and continues streaming progress events.
Mapping status and session management
Each mapping task runs within a session. You can check the status of a session at any time to see overall progress, how many target fields have been mapped, and how many decisions are still pending. If a mapping task is no longer needed, a cancellation request can be sent to stop an active stream for that session.
Uploaded context files
Context files you upload for a session are stored and associated with that session. You can list the files attached to a session and download any of them again if needed. Uploaded files are retained for a limited time and may expire.
Idempotent requests
If you submit the same mapping request more than once using the same idempotency key, the platform recognizes the duplicate and returns the original result rather than starting a new task. If the same key is reused with different parameters, the platform returns an error so you are aware of the conflict. Idempotency keys must be no longer than 255 characters.