Building an AI-ready, cost-controlled open source observability platform: Escaping SaaS bloat

In an AI-driven world, observability costs can spiral out of control. In this session, Kishan Baranwal shares how his team replaced an expensive commercial SaaS tool with a right-sized, optimized, GitOps-friendly open-source platform built on OpenTelemetry. Rather than rebuilding a fragmented stack, the team designed observability as an internal product: multi-cluster by default, tenant-isolated, cost-aware through intelligent shaping, and ready for AI-assisted debugging. The talk covers key architecture decisions, a low-risk migration strategy, and packaging tactics that drove rapid adoption.

Platform principles:
• Multi-cluster first: clusters as domain primitives
• Per-cluster ingestion: no central choke points
• Everything-as-code: GitOps as a single source of truth
• Correlation by default
• Isolation guaranteed
• Open-source system of record

Key takeaways:
• Safe blueprint for migrating from SaaS APM
• Designing scalable multi-tenant observability
• Cost controls that preserve full debuggability
• Packaging observability for real adoption
• Preparing telemetry for AI-driven operations