Signed in ·
Signed in ·
Check for event details.
Defaulting to cloud AI APIs leads to spiraling bills and compliance risks. Rushing a local migration with a day-one terraform apply on cloud GPUs is just as dangerous. Without a proper System Design Phase (SDP), teams over-provision infrastructure they don't understand. This talk outlines "Sprint 0" for self-hosted, Sovereign AI. Learn to run PoCs and document ADRs to calculate VRAM capacity for open-weight models (DeepSeek, Qwen) before provisioning. We dive into a GitOps-driven Kubernetes architecture using ArgoCD. Learn how we balance workloads across datacenter-grade H100s via vLLM and local RTX workstations via Ollama to respect NVIDIA’s EULA constraints. Finally, we cover the FinOps and security layer: using LiteLLM for token tracking, Kong for rate-limiting, and Istio mTLS to secure proprietary data inside your network.
Format: USE CASE
Location: ROOM C
Speakers: Milenko Mitrovic
Level: Intermediate to Advanced
Audience: Registered participants