| Implement generative AI and agentic solutions | Deploying and consuming LLMs, small models, code models and multimodal models. Implementing retrieval-augmented generation in an application. Designing tool-augmented flows and multistep reasoning pipelines. Evaluating apps for fabrications, relevance, quality and safety. Building agents: defining roles, goals, conversation tracking and tool schemas, function calling, conversation memory, orchestrated multi-agent solutions, and autonomous workflows with approval controls. Tuning generation behavior, including prompt engineering. The largest domain. | 30 to 35% |
| Plan and manage an Azure AI solution | Choosing the right model for a task across LLMs, small language models and multimodal models. Choosing Foundry services for generation, grounding, vector search, agent workflows. Choosing retrieval and indexing methods. Designing infrastructure and deployment options, and wiring Foundry projects into CI/CD pipelines. Managing quotas, scaling, rate limits and cost footprints. Monitoring drift, safety events and grounding quality. Security via managed identity, private networking and keyless credentials. Responsible AI: safety filters, guardrails, evaluators, trace logging, provenance metadata, and governing agent behavior with oversight modes. | 25 to 30% |
| Implement computer vision solutions | Vision capability built through Foundry rather than standalone Cognitive Services endpoints, including multimodal interpretation of visual input and generative image output. | 10 to 15% |
| Implement text analysis solutions | Keyword extraction, entity detection, sentiment analysis and summarization, plus speech, delivered through Foundry Tools. | 10 to 15% |
| Implement information extraction solutions | Extracting structured information from documents, forms, images, audio and video using Azure Content Understanding. This is the successor to a lot of what used to be Document Intelligence and Form Recognizer work. | 10 to 15% |