At Semrush, I owned a shared Python library and extended the data-quality pipeline that checks external suppliers before their data enters production. I reconciled dependencies across three repositories and helped get their CI/CD pipelines green end to end.
I also migrated the traffic model’s training data from GA3 to GA4 using a gradual blend validated on one country before rollout. For the model, I built a URL classifier for identifying completed purchases and delivered scored results with a written analysis.
On HF Model Finder, I built a RAG search over Hugging Face model cards, with an evaluation harness to compare retrieval and answer-generation choices. The project includes Docker Compose and Kubernetes deployments, tests, and CI.
