You're working on a product where a large language model needs to complete multi-step tasks, use external tools, and ground answers in trusted knowledge instead of relying only on its pretraining. Some workflows are simple single-turn requests, while others require retrieval, planning, and tool use.
How would you approach a project involving large language models and agent-based systems?
Using LLMs for grounded task completionDesigning agent workflows with tools and retrievalEvaluating answer quality and agent behaviorHandling prompt injection and unsafe tool use