redis-developer/redis-nvidia-recsys

Implements three progressively sophisticated architectures—offline batch processing with Two-Tower models, low-latency online serving via Triton inference with Redis feature stores, and distributed HugeCTR training for billion-scale embeddings. Uses Docker/Compose for local GPU execution and includes Terraform/Ansible for AWS production deployment with Redis Enterprise, monitoring, and VPN infrastructure. Demonstrates the full pipeline from feature engineering with NVTabular through real-time inference, bridging batch preprocessing and sub-300ms online recommendations.

No commits in the last 6 months.