RAG: The 2025 Best-Practice Stack, Prototype to Production

Value of the Information & Strength of the Argument

The video offers significant practical value by presenting a concrete, actionable stack for RAG development, based on the speakers’ extensive experience in AI engineering education and consulting. The argumentation is solid, as they justify each tool choice with clear reasoning, such as LangGraph’s flexibility for complex systems, Qdrant’s scalability, and RAGAS’s credibility in evaluation. They also provide a realistic phased approach to production, acknowledging the challenges of enterprise adoption. However, the content is primarily opinion-based, lacking formal citations or comparative benchmarks, which limits its scientific rigor.

The title accurately reflects the content, which focuses on the best-practice RAG stack for 2025 and the journey from prototype to production.

7/10

The video presents a well-structured, practical overview of a recommended RAG stack, drawing on the speakers' extensive industry experience. Claims are substantiated with reasoning and comparisons, but the content is largely opinion-based and lacks formal citations or empirical validation.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 00:00Introduction and team presentation
• 05:00Overview of RAG concepts and the importance of fact-checking
• 10:00Discussion of the best-practice stack: LangGraph, Qdrant, Cohere Rerank, RAGAS, Together AI, Llama 3.3
• 15:00Explanation of the five phases from prototype to production
• 20:00Live coding session: building a simple RAG application
• 30:00Deep dive into evaluation with RAGAS and monitoring with LangSmith
• 40:00Discussion on cloud vs. on-premise deployment and enterprise considerations
• 50:00Q&A session and community engagement

• AI Engineering Bootcamp — Mentioned as a resource for further learning and the source of their tested stack.

• A16Z LLM Stack — Referenced as the foundational architecture for LLM applications.