A major challenge in making AI accessible is the engineering which it takes. We’ll discuss practical aspects of implementing LLMs in software systems and running the model-evaluate-deploy loop. Topics will include fine-tuning, retrieval-augmented generation, and software system challenges. We'll also look at ways to reduce complexity and cost. The talk will provide insights into setting up secure and robust systems, ensuring machine-readable and non-toxic outputs, and maintaining a standardized output format. We’ll end with open research questions which are still interesting to industry and practitioners.