Scaling AI From Prototype to Production With Open Source Databases

AI scaling has become a major concern for enterprises, with AI projects often failing after the prototype stage. In this episode of Enterprise Unlocked, pgEdge co-founder and Chief Product Officer (CPO) Phillip Merrick, discusses the major reasons for this gap and lays out solutions that focus on the underlying data infrastructure. Structured data is central to the success of AI projects, as it determines the quality of output AI models deliver. If not provided with the right data, AI tools might not deliver the desired results and even hallucinate. Against this backdrop, Phillip discusses how database decisions impact the success of AI projects and the security and governance considerations that go into these decisions. Postgres and other open source databases make for flexible infrastructure that can be tailored to organizations’ unique requirements. He also discusses the idea of giving back to the developer community and how open source fits into this, while offering key advice for developers. “Design with the end in mind. If this is not going to be a throwaway prototype, if it is something potentially you want to take to production, then understand what it will take ahead of time to get it to production.” ~ Phillip Merrick, Co-founder and CPO of pgEdge Key Questions Answered: • Phillip, you have experienced multiple technology shifts in your career. Tell us about the learnings that will help developers today. • How can developers upskill themselves? How do they work with probabilistic systems well? • What are some other practical strategies that can help developers bridge this gap we've been talking about between prototype and production? • Does the strength of the underlying data infrastructure impact the success of AI projects, especially in the enterprise context? • How do database decisions determine whether AI projects scale, especially enterprise AI projects? • What's the role of open-source databases, particularly Postgres, in creating successful agentic AI systems? • What are some overall guidelines you'd suggest to enterprises to help with successful AI deployments, maybe not just in development but even across departments? How should the entire cycle work, just as an overview? Key Themes and Takeaways From the Conversation: 1. Gap Between Enterprise AI Prototype and Production • Many AI models do not move from prototype to production • There is a need for compliant and secure infrastructure for AI applications to run • Not all enterprise data can be outsourced to external cloud environments 2. Deterministic vs Probabilistic Systems • AI has shifted the responses developers get from being deterministic to being probabilistic • Systems can now give responses on the line of: ‘it depends’ • To deal with the shift in probability, chain a series of agents together instead of doing it monolithically 3. Importance of Data Infrastructure • Structured data is the ground truth for AI • Lack of proper data can lead to AI hallucinations • Good prompts with clear data location can be useful 4. Postgres and Open Source Databases • Postgres is managed by a community, and not tied to a company • It is fully open source with an OSI-approved license • There is no vendor lock-in associated • It builds the idea of giving back in the software community 5. Practical Advice for Developers • Design with the end in mind • Security, compliance and governance requirements need to be clear • Be clear about the plan for production, design and getting the prototype to compliant environment Chapters: 00:00 Introduction 02:15 Prototype to Production Challenges 08:25 Infrastructure Needs for AI Success 14:45 Lessons from Web Services History 21:30 Importance of Structured Data in AI 28:10 Open Source Contributions and Impact 35:00 Strategies for Successful AI Deployment 42:00 Conclusion and Thank You