Applied AI @ rumble.com (contract)
Rumble is a video sharing platform with over 100 million active monthly users, founded in 2013.
Leading modernization of ML/AI systems across discovery, ranking, recommendations, and real-time personalization. Took ownership of underperforming efforts, rebuilt them from the ground up, and delivered measurable value for users and the business.
A hands-on technical leadership role, architecting and delivering scalable ML/AI systems that drive discovery, ranking, recommendations, and personalized experiences across the video sharing platform.
- Data: Revamp data engineering into a continuous pipeline from source data through derived data for various target platforms.
- Behavioral: Improved the precision and utility of user behavior tracking, aligning it with the needs of downstream pipelines.
- Scalability: Designed a multi-tier data architecture to balance long-tail retention, emerging trends, and short-session real-time response. Tuned GPU usage across local and remote resources for utilization and cost-efficiency.
- Technology: Strengthened the data platform’s Kubernetes architecture, added feature and vector stores, introduced CDC and durable event flow, enhanced OpenSearch for text and vector queries, and integrated NVIDIA Merlin for recsys workloads.
- Observability: Built deep observability into all stages of the pipeline, monitoring freshness, transformation, and drift.
Open-source used within our technology stack includes:
Technology stack centered around:
- OpenSearch
- Qdrant
- FAISS
- NVIDIA Merlin
- Hugging Face Transformers
- PyTorch
- Redis / KeyDB
- MySQL
- Clickhouse & StarRocks
- Parquet & Iceberg
- Kubernetes
- MinIO
- Kafka / Redpanda
- Apache Spark
- Ray
- VictoriaMetrics & Logs
- Python
- Kotlin w/Quarkus