Admir Krnic

At Qyobo, I own regulatory and supply-chain data domains end to end for a pharmaceutical market-analytics platform. I turn unstructured, contradictory sources into reconciled product outputs through SQL and Python pipelines processing petabytes per sync and billions of rows.

I rebuilt drug-data parsing to achieve 70–100% extraction success across covered sources, expanded regulatory coverage by 10+ countries, and redesigned authorization data to approach 100% manufacturer, pack, and substance mapping accuracy in priority markets.

Alongside industry work, I’m a first author researching neural audio watermarking at DeepMark. My work formalized backward compatibility in zero-bit watermarking, restoring true-negative rates to 99–100% across out-of-chain embedders, and improved robustness to 95–99% bitwise recovery across 25 distortion and attack types.

I’m strongest when the data is ambiguous, poorly documented, and lacks clean ground truth. I work across Python, SQL, Postgres, Trino, dbt, SQLMesh, Temporal, PyTorch, and AWS.