• Led the upgrade of the clusters backing Playlist-Delivery, the public API serving playlists to our players around the world, from ElasticSearch 2.4 to ElasticSearch 7.16. It spanned over two quarters, involved five teams and impacted seven different services. It was completed with no disruption to customers.

• Evangelized Alpaca, our in-house ETL platform built on top of Airflow, among the engineering organization. It now runs 330 DAGs.

• Maintained and added features to Playlist-Delivery, the public API serving playlists to our players around the world. At the time, it handled about 110 million unique requests a day with a sub-500ms latency, and was backed by a set of ElasticSearch clusters.

• Collaborated in the upkeep and performance tuning of the above-mentioned ElasticSearch clusters.

• Participated in the development of Alpaca, our in-house ETL platform built on top of Airflow. It then orchestrated about 100 individual DAGs, running as frequently as every 20 minutes.

• Collaborates to the development of the company’s public reporting API, surfacing performance data via a JSON-based query language. Requests would be serviced in the back-end by various databases (Postgres, Redis, MongoDB, Elasticsearch, MySQL, S3, Snowflake ...). The API enables publishers to better understand and target their audience.

• Participates in the ideation and implementation of labs, which are PoCs pushed to publishers through a designated page on the company’s dashboard. A lab gathering significant positive feedback would then graduate to full-on production service and be fine-tuned to incorporate publishers’ comments and suggestions.

• Cooperates in the design, coding and maintenance of ETL solutions to support easy internal reporting.

• Designed a collection of Spark Java applications, streaming from RabbitMQ and Kafka clusters, to monitor business KPIs in real time. Each detector follows the multi-dimensional evolution of a specific metric, compares it with pre-calculated forecasts, and pushes noteworthy JSON-formatted anomalies to a single web service.

• Created a RESTful Java web service that combines noteworthy information from KPI-specific Spark detectors and sends actionable/curated alert notifications to account managers via e-mail and to a dedicated UI via POST requests.

• Streamlined Bing's data consolidation, performance analysis and reporting. Involved automating performance-data retrieval from the PubCenter reporting API, reconciling their records with ours, running multi-dimensional analysis and providing a clear overview to stakeholders.

• Upgraded internal audit of our CPA/ROAS performance on direct advertisers. This software now acts as a valuable feedback loop for account managers to identify jeopardized accounts, assess benefits from their optimizations and pitch for budget increases.

• Modeled a heat-map generator to offer an impactful, intuitive representation of traffic overlap between publishers. Identifying real traffic sources from brokers prevents us from competing against ourselves in auctions and artificially drive CPCs up.

• Revamped internal publisher scoring to better assess the quality of traffic sent by publishers. Trimmed run-time down by 75%.

• Automated generation of existing traffic quality reports, which fastened crafting process by 90%. Proposed new metrics and designed new reports in conjunction with Senior Management.

For both these projects, applications were written in Python, SQL and VBA.