FareedKhan-dev/scalable-rag-pipeline

Implements multi-step agentic reasoning through LangGraph state machines (Planner, Query Rewriter, HyDE), combining vector search via BGE-M3 embeddings in Qdrant with Neo4j graph queries for structural knowledge. Deployed as a decoupled two-plane architecture—CPU-based Control Plane handling orchestration and GPU Data Plane using Ray Serve for LLM inference (vLLM) and embeddings, with Karpenter-driven autoscaling on AWS EKS and Terraform IaC for reproducible provisioning.

111 stars.