The session examined the practical lessons and recurring pitfalls encountered while launching generative AI products. It separated hype from measurable business value and treated AI development as research and development rather than predictable feature delivery.
What the session covered
- The difference between generative AI and traditional AI systems
- Business value, success metrics, and the limits of impressive demos
- Naive and advanced RAG patterns, metadata, reranking, and evaluation
- LLM quality, prompts, source data, hidden data-acquisition costs, and fine-tuning
- Open-source models, infrastructure requirements, and the cost of operating model workloads
- Four levels of generative AI systems and the production challenges at each level