Media Search Query Classifier: ML Pipeline for Mediascope
A hackathon ML project for classifying noisy user search queries using the TypeQuery, Title, and ContentType fields. The solution identifies whether a query refers to professional video content, normalizes the title or franchise, and predicts the content type, including films, TV series, animated films, and animated series.
I was responsible for the final integration of the ML pipeline: improving inference, stabilizing the solution, preparing model artifacts, and combining signals from multiple components into the final prediction. The pipeline used dictionary matching, retrieval, embedding-based similarity, graph features, and prediction calibration.
A team hackathon project that placed third at AI Business SPB. Repository: GitHub