Marc Kelechava Email & Phone Number

California, United States

Punchup is developing innovative tools to allow live entertainers to own and activate the relationship with their fans- Feed Ranking models- Recommendation models- 0 to 1 Snowflake DE Infra- Nextjs web work- Experimentation

San Francisco, Ca, Us

Building Ads Bidding & Auction from 0 to 1 :)

San Francisco, Ca, Us

Designed and implemented interleaving ranking measurement BE and data side end-to-end enabling 7x faster online testing decisionsDeveloped an Airflow/Sagemaker API ranking model retraining framework that reduced training set creation time from multiple days to just 1 hour, resulting in a significant increase in model training velocity and experiment launches.Implemented a solution to monitor production real-time ranking models, enabling the tracking of long-term ranking trends at the surface level and alerting on anomalies or model degradation.Developed and implemented a frontend viewport impression logging solution, improving impressions data accuracy and increasing retailer engagement across platforms and surfaces. Integrated new data into critical business processes, used in successful experiments on global GMV and order lift

San Francisco, Ca, Us

Led successful migration of 75% of Discovery team’s DAGs/Python ML Apps from Redshift to Snowflake, including the entire ML Features Store, with no downtime and no failures.Modularized existing product impression attribution DAGs for ranking surfaces and refined cart/order attribution to explicitly define training labels, resulting in easier custom label engineering. Used results for a suite of surface-specific metrics used in many dozens of A/B experiments, and tables used downstream by 100s of internal use cases

Sebastopol, Ca, Us

-Implemented a user-conversion classifier for O'Reilly's online subscription product team to uncover most impactful platform usage features.-NLP work creating embeddings on raw full text epub content to map different units to a vast subtopic hierarchy, then study the related usage

New York, Ny, Us

Went through the Metis program: a rigorous data science bootcamp providing training in machine learning and multiple programming languages.

Produced many tracks, performed live many times.

-Collaborated with external client data science teams to shift ad spend from other marketing channels (e.g. TV ads). Shifted an estimated $100+ Million in ad spend to FB from other channels for very large clients.-Co-led a custom ad targeting initiative for large US ad clients (writing SQL/Hive DAGS on huge amounts of usage data) that generated > $25MM in half of 2014.-Conducted quantitative studies of Facebook usage and engagement from a vertical perspective e.g. analyzed global ad sentiment data via SQL and Hive pipelines & showcased with Tableau visualizations.-Built scalable dashboards and data pipelines for team members to extract insights for bidding and optimal media buying

New York, Ny, Us

Jersey City, New Jersey, Us

Cambridge, Ma, Us