Stop Rebuilding Your AI Pipelines: The Hidden 90% of Production AI Infrastructure - Maher Hanafi

Stop Rebuilding Your Pipelines: Scalable Data Architecture Patterns for Production AI Most AI teams focus on models. The teams that succeed in production focus on infrastructure. In this talk from the Data Infrastructure Summit 2026, Maher Hanafi (SVP of Engineering at Betterworks) shares lessons learned building and operating AI systems at scale—including the reliability failures, architectural challenges, and hidden infrastructure requirements that emerge after the first successful launch. The talk explores why production AI is far more than model selection and prompt engineering, and why the real work happens in the layers beneath the surface: reliability, evaluation, observability, security, retrieval, governance, and operational feedback loops. Topics covered: • Why Betterworks chose an inference-only, model-agnostic strategy • The correct optimization sequence for production AI systems • Reliability lessons from real-world deployments • Offline inference and pre-processing strategies • In-flight batching and GPU utilization optimization • Building Responsible AI guardrails without sacrificing latency • Prompt management as production infrastructure • Custom evaluation pipelines and quality gates • RBAC-aware retrieval in multi-tenant systems • Designing AI platforms that work across both monoliths and microservices • The hidden 90% of AI architecture that users never see If you're building AI products, GenAI platforms, RAG systems, agentic workflows, or enterprise AI infrastructure, this session provides practical architectural patterns that can help you avoid common reliability, security, and scaling pitfalls. Speaker: Maher Hanafi SVP of Engineering, Betterworks Event: Data Infrastructure Summit 2026 #AIEngineering #GenAI #MachineLearning #DataEngineering #MLOps #RAG #LLM #SoftwareArchitecture #PlatformEngineering #AIInfrastructure #ArtificialIntelligence #EnterpriseAI