Turning healthcare data transformations into checked outputs

The problem I was working through

At Cedar Gate Technologies, the work was not just about moving healthcare data from one shape to another. The harder question was whether the transformed output still matched the client’s expectations closely enough to be delivered.

I worked as an Associate Data Engineer from January to October 2023 in Lalitpur, Nepal. The role combined analysis, transformation, validation, and collaboration with teams delivering work for dedicated clients.

Where I focused

I analyzed healthcare datasets using Amazon Redshift, PostgreSQL, and SQL, with statistical techniques including random sampling, normalization, and standardization. For transformation work, I used the organization’s proprietary File Mapper and Scrub engines.

The useful lesson was that each stage answered a different question:

• Analysis helped understand the incoming data.
• Transformation made the data fit the required output structure.
• Validation checked whether the output still matched expected controls.
• Collaboration made sure technical outputs aligned with client requirements.

Workflow overview

The diagram summarizes the work without exposing client datasets, schemas, or proprietary engine internals.

What the metrics mean

The pipelines processed 200 GB+ of raw client data each week. At that scale, small assumptions can become repeated errors, so the validation step mattered as much as the transformation step.

The reported outcome includes 99% data integrity before delivery. I treat that as the integrity measure reported for this work, not a universal definition of healthcare data accuracy. The important point is the operating habit behind the number: a completed pipeline run is not enough by itself. The output still has to be checked against agreed controls.

The engineering habit this reinforced

This work made data quality feel practical rather than abstract. A pipeline can finish successfully while still producing an output that needs review. A transformation can be syntactically valid while still shifting the meaning, distribution, or usability of the data.

That distinction carries into my current platform and data work. Infrastructure provides a place for jobs and applications to run; validation decides whether the result is useful enough to trust.

Sources and context

• My engineering toolkit shows how the technologies across selected projects connect.