Overview

Google, IBM & Meta Certificates — All 10,000+ Courses at 40% Off

One annual plan covers every course and certificate on Coursera. 40% off for a limited time.

Demonstrate you have the job-ready skills to design and implement a complete AI system from data to deployment, with this portfolio-worthy RAG and Agentic AI Capstone Project from IBM. You’ll design and build a production-style multimodal RAG system that combines structured data, embeddings, retrieval logic, evaluation strategies, and intelligent workflows into one cohesive, scalable solution. You’ll create and manage structured JSON datasets, generate text and image embeddings, and construct a vector database to power accurate similarity search and metadata-filtered retrieval. As you progress, you’ll implement robust RAG pipelines, apply re-ranking and evaluation techniques, and strengthen response quality using multimodal inputs and systematic validation approaches. You’ll also design a multi-agent recommendation system, integrate tools using the Model Context Protocol (MCP), orchestrate workflow testing, and launch an interactive Gradio chatbot interface. By the end, you’ll have developed an end-to-end generative AI application that demonstrates practical AI engineering expertise, architectural thinking, and production-ready implementation skills.

Syllabus

Module 1: Build a Structured Generative AI Application

In this module, you will use LLMs to transform unstructured restaurant descriptions into structured JSON files by designing prompts and extracting predefined attributes. You will apply multimodal LLMs to generate captions from review images and integrate those captions into structured user review data. Finally, you will build a command-line Python interface to browse, add, edit, and delete restaurant records, integrate LLM-powered structuring functions for new entries, and implement file backup mechanisms before saving updates.

Module 2: Design a Multimodal RAG System

In this module, you will design and implement the retrieval layer of a multimodal RAG system using structured restaurant text data and food images. You will construct multimodal vector indexes, generate text and image embeddings, and build retrieval workflows that combine similarity search with metadata filtering. You will also implement late-fusion techniques to combine and rerank results across modalities, improving the relevance of retrieved outputs. The module follows a step-by-step retrieval pipeline, from index construction to hybrid retrieval and multimodal ranking, with a focus on practical design rather than tool-specific features.

Module 3: Combine Agents into a Multi-Agent System

In this module, you will design and implement a multi-agent recommendation system. You will define specialized agents with clear roles, goals, backstories, and tasks, and integrate them into a coordinated multi-agent workflow. You will then test how multiple agents collaborate to generate restaurant and recipe recommendations from a single user input. Finally, you will build an interactive chatbot interface using Gradio to expose the system. The chatbot will process user queries, display coordinated agent outputs, and support basic database editing functionality within the interface.

Module 4: Integrate Agents, RAG, and Tools with MCP

In this module, you will organize agent tools, databases, and documents within an MCP server. You will then build an MCP client and an LLM-based MCP host that communicate with the server and validate the system through testing. You will also design and implement an LLM-powered MCP host with a GUI, enabling the LLM to access server-exposed tools and documents. This module brings together components built earlier into a unified MCP-based system and validates end-to-end tool execution through a GUI-based application.

Module 5: Final Project and Course Wrap-Up

In this module, you will complete your AI capstone project by submitting screenshots of tasks performed in previous labs. You’ll organize and present these artifacts to clearly demonstrate how you designed, built, and integrated structured data, multimodal RAG systems, and multi-agent workflows using LangChain, LangGraph, and MCP. This submission will serve as a final evaluation through an AI-based grading system and provide a portfolio-ready showcase of your end-to-end generative AI solution.