Accelera

ETL automation tool focused on data migration. Removes the manual lift from moving and transforming data between systems — the repeatable parts that today still consume engineering quarters.

Data migrations between systems still take engineering quarters of work, mostly in writing throwaway transformation code that gets thrown out the moment the migration is done.

Most migration work is repeatable but tooling treats every migration as bespoke. Automating the repeatable parts unlocks weeks of engineering time per project — and removes a category of work nobody wants to own.

Connector model. datasource_connector/ provides a pluggable connector framework over a base connector.py abstraction — apiConnector/ (REST + token auth), s3Connector/ (boto3), kafka_connector.py, rabbitmq_connector.py, and DB connectors for Postgres/MSSQL/MongoDB. Packaged as a standalone SDK (dsconnectors in accelera_sdk_poc/) so external services consume it without touching core code. Transformation engine. comply_backend/app/ — data_transformer.py + data_enrichment_service.py handle field-level ops (pad, trim, concat, mask, type cast), with mapper/ holding mapping configs. Pipelines are stored as JSON in dbo.t_pipeline_config and parsed at runtime (pipeline.py:33), which enables no-code pipeline editing via the UI. Scheduling. comply_backend/velo.py uses APScheduler's BackgroundScheduler for interval-based jobs; Kafka (AWS MSK) handles async event streaming between ingestion and transformation. Validation + reconciliation. app/utils/max_rules.py encodes SWIFT MT103 rules — BIC codes, account hierarchies, intermediary agent fields — and date_validation.py covers temporal checks. transformation_routes.py tracks transformation status with createdBy/updatedBy audit fields and project-scoped status checks across multiple flows.

Fintech ETL taught me that the technical problem is rarely the hardest one — coordinating people across timezones, formats, and competing priorities is. I learned to push back gently when scope drifted, to give the team space to think before reacting to a fire, and that empathy with the ops folks running the pipelines is what separates tools people tolerate from tools they defend. The stack matters; how you treat the people running it matters more.