Agentic RAG on Kubeflow
summary
Project Flow implements an agentic RAG pipeline on Kubeflow that runs on KServe + vLLM for model serving, uses Milvus vector DB and Kubeflow Pipelines (cron indexing) for ingestion and updates, and exposes persona-based agents via a KAgent front that calls MCP services. It self-hosts a self-hosted 2.5–7B model, tracks per-query token usage, creates 30-minute JWT sessions, and pilots a chat widget on the site; goal is production-grade reproducibility, evaluation, and security.
AI-written summary · about
chapters (9)
- 0:00Project Flow on Kubeflow
- 1:25Self-host LLMs for security and cost
- 2:49Agentic RAG with KServe and Milvus
- 5:39Building reference agentic RAG
- 7:24Querying GitHub docs and YAML
- 9:10Evaluating and improving RAG
- 10:13Prototype to production
- 12:21Demo: Flow chatbot pilot
- 14:04Q&A
related talks
36mShift-Left for LLMs: Securing the AI Model Supply Chain
31mKubeflow 2026 State of the Union: Orchestrating the Full ML Lifecycle
41m“Hey AI, Train Llama”: Making Kubeflow Agent-Native With MCP
27mKubeflow for enabling AI Powered Drug Discovery and development...
31mKubeflow - Data Science on Steroids
14mKubeflow: Where ML Meets Kubernetes
44mEngaging the Kubeflow Community: Building an Enterprise-Ready AI/ML Platform
34mKubeflow: Multi-Tenant, Self-Serve, Accelerated Platform for Practitioners