AI-powered ETL and scheduling migration
Export in, assessment out: what converts as-is, what needs a hand, what to retire. Deterministic conversion first, an agent for the long tail, a human review on every diff. Parity against the old estate is the acceptance test, not a promise.
Where you are
The licence renews every year because leaving is impractical: thousands of mappings and jobs nobody fully understands, written by people who have left, with the business logic buried in transformations and schedule conditions. Vendor tools convert to that vendor's engine only. Consultancies quote by the mapping and by the month.
How the engagement runs
Assessment from the export
You export the repository (powrmart XML for PowerCenter, the folder export for Control-M). We never need access to the server. The assessment classifies every mapping and job: converts as-is, converts with listed TODOs, or needs a person, with the reason for each. The estate's transformation types, features and workflow completeness are reported, with the dead code and the missing references.
Deterministic conversion
Mappings compile to dbt models: one model per target, sources typed from the export, key tests generated, Informatica expressions translated to dbt cross-database macros so your adapter renders the SQL for Snowflake, Databricks, Athena or ClickHouse. Workflows and job folders become Airflow DAGs: sessions run dbt, commands run as Bash, links and conditions become dependencies.
An agent for the long tail
Where the compiler flags a TODO, a Claude-based agent with read access to the mapping definition proposes a resolution, validates the SQL and submits it. Nothing it writes replaces what the compiler was sure of, and every change is a reviewable diff.
Parity as acceptance
Old and new run side by side on the same inputs; row counts and checksums per target decide. A mapping is done when the numbers match, not when the code looks right.
Already built
- ●Control-M → Airflow converter: 30 job types, tested against a real 499-folder estate export, 127 automated tests.
- ●Informatica → dbt converter: parser, assessment report, expression translator, row-stream compiler, dbt project writer, workflow → Airflow renderer and the Tier 2 agent, verified by a real dbt build of the generated project.
- ●Both converters are open source under Apache-2.0; the assessment runs on your laptop and nothing leaves your machine.
Normalizer, Java and Stored Procedure transformations, mapplet flattening and SQL-override rewriting are flagged for a person rather than guessed. Session-level overrides and pushdown are out of scope.
What you bring
- ●A repository export (read-only; no server access needed)
- ●Your target engine and dbt adapter
- ●An Airflow, or let us stand one up
- ●One engineer who knows the estate, for the review queue