See how agents call tools by wrapping functions or services and deciding when to use them. Use web search, file search, and code interpreters to ground responses.

Explore agentic autonomy with tools and context engineering and management to enable autonomous tool selection, precise input schemas, and clear workflows that chain tools to achieve complex goals.

Explore retrieval-augmented generation, grounding llms with external data stores and vector stores to curb hallucinations, and learn how an agent uses rag as a tool for fast, cost-efficient queries.

Learn embedding vectors and vector stores for retrieval-augmented generation. Compare vector databases like Chroma DB, Elasticsearch, and Neo4j, and master chunking and querying strategies.

Follow along as we set up a rag vector store with chroma db, read the calories CSV, convert rows to documents with metadata, and populate a nutrition collection for retrieval.

Extend the tool calling function to connect the RAC database and run calorie queries against the nutrition DB, using RAG-style retrieval for top results.

Build your own retrieval-augmented generation setup by connecting the nutrition Q&A rag database to an agent, using two tools to retrieve malnutrition and pregnancy data.

Discover how model context protocol (MCP) standardizes retrieving context from external systems, exposing tools, and connecting agents through local or cloud MCP servers and clients.

Learn to implement tool calling with the exa search MCP by configuring an MCP server and caching, while using gallery lookup and web search for ingredients and calories.

Replace the MCP tool with OpenAI's built-in web search tool, a simpler, managed option. It costs a small amount; use the solutions notebook for a peek, and complete the exercise.

Explore short-term and long-term memory in AI agents, learning how session objects store chat history and extracted insights, plus options to persist memory with SQLite, SQLAlchemy, and back-end databases.

Explore hands-on agentic memory by implementing a persistent short-term memory with OpenAI sessions using a sqlite-backed session to let an agent remember apples, bananas, and lollipops across runs.

secure ai agents by enforcing prompt adherence and explicit tool schemas, use few-shot examples and evaluation tests, and guard against leakage, prompt injection, and cross-session data exposure.

Guardrails protect input and output from unwanted content, filter prompts, keep agent outputs on target, and run in parallel to agents via tripwire exceptions.

Defend production LLM chatbots from jailbreaks with guardrails and input-output checks. Build guardrail agents using system prompts and tripwire logic to keep conversations on topic and prevent prompt leak.

Implement secure deployment practices: use oauth for access, rotate api keys, isolate internal calls, minimize pii with guardrails, monitor budgets, and consider smaller models plus tool-based reasoning to control costs.

Build a production-grade chatbot with Chainlink in Python, featuring a simple user interface and an echo-style handler, in the chatbot folder with a logo and port 8000, without MLM functionality.

Develop an agentic chatbot with Python and OpenAI, featuring a nutrition agent, Chroma integration, streaming tool calls, and rag capabilities. Extend with memory, authentication, and deployment for a multi-agent system.

Learn how to manage conversational memory with OpenAI sessions using Chainlink, initialize SQLite conversation history and user/agent sessions, and persist context across chat runs.

Learn a practical approach to adding authentication to an AI agent app by implementing username and password protection, managing secrets via environment variables, and deploying a protected endpoint for production.

Commit and push changes to GitHub, then deploy the chatbot to production on render as a Python web service. Configure port 10000, environment variables, and verify the login flow.

Implement guardrails and multi-agent execution in the chatbot and deploy it to a render endpoint, building on the nutrition agent in this capstone project.

Explore OpenAI’s agent kit and agent builder to create visual, drag-and-drop agent workflows with nodes, guardrails, tools, and evaluations, then preview, debug, and deploy.

Build simple agents in the agent builder UI by defining agents and adding canvas nodes; configure chat history and the GPT five nano model with logic and data transformation.

Build a nutrition agent that uses OpenAI's rag vector store via the file search tool, loading the gallery database and Q&A datasets to answer calorie-related questions.

Divide the workflow into an ingredients agent and a calorie agent using web search and a custom MCP server to produce a readable meal calorie summary.

Learn how to use Agent Builder guardrails to enforce preset checks, prevent personal data exposure, enable moderations, safeguard against jailbreaks, and reduce hallucinations by linking to a vector store.

Implement guardrails and structured messages in AgentBuilder to constrain an agent to nutrition topics, using a topic classifier output, if-else logic, and clear error reporting for off-topic prompts.

Learn how to evaluate AI agents with evals and trace graders in the OpenAI agent kit, using LLMs as judges, data sets, and prompt optimization via the evals API.

Learn to evaluate AI agents with an LM as judge, using graders to prevent regressions and enforce guardrails against reference inputs in a simulated CI/CD workflow.

Publish and deploy your agent with ChatKit by embedding customizable UI widgets in your app, using OpenAI hosting or your own backend, with token-based authentication.

Publish and deploy your OpenAI chat workflow, then integrate the agent on your website via the Python SDK or Chat Kit, configuring env files for API access.

Extend the multi-agent system by adding a sports coach that creates an exercise plan. Retain ingredients and coloring agents and update topic classifier and condition block for exercise plan requests.

Celebrate completing the AI agents crash course by downloading your Udemy certificate, sharing it, and connecting with the instructors on LinkedIn; consider leaving a detailed review.