Most AI systems still expect clean text. The work people need done arrives as screenshots, scanned invoices, product photos, voice notes, and charts buried in PDFs. This Specialization covers multimodal AI agents: systems that interpret all of it, then act on what they find.
You start with what separates an agent from an assistant, then add prompt design, memory, and tool calling until it does something useful. You bring in vision-language models for multimodal search, coordinate agents on one task, then test, guard, and ship what you built.
By the end of this Specialization, you will be able to:
1. Describe the agent lifecycle from perception through reasoning to action.
2. Process text, image, audio, video, and document input.
3. Implement tool calling with reliable schemas and error handling.
4. Build multimodal retrieval over documents containing charts and tables.
5. Coordinate multiple agents through defined roles and shared context.
6. Test, guard, and deploy agents with REST APIs and containers.
This Specialization suits AI engineers, machine learning engineers, applied AI developers, automation engineers, and solution architects moving past prompt-based chatbots into systems that act. It assumes basic Python and comfort calling APIs, and no background in agents, computer vision, or speech processing.
Enroll now to build multimodal agents that see, listen, reason, and act.
Projet d'apprentissage appliqué
Across the Specialization, you complete hands-on projects that mirror the work of building agents for real inputs rather than clean text. You build an agent that accepts combined text and image input, calls an external tool, and uses stored history to answer a follow-up question. You then construct a multimodal retrieval system that answers questions about charts inside PDF documents, and a pipeline in which two specialized agents hand off context to each other.
In the final project you deploy a multimodal assistant with a Streamlit interface, add input and output guardrails, log decision traces, and measure latency and token cost across model calls. Each project produces a concrete artifact, including working agents, tool integrations, guardrail configurations, and a deployed interface, that you can add to a professional portfolio as evidence of practical skill.
