Build a Custom AI Agent for Your Field
Build a domain AI agent with the Claude/OpenAI API and a knowledge base (RAG) that automates a real workflow in your discipline.
About this course
You'll build a custom domain AI agent — a research assistant, an intake bot, or a discipline-specific helper — using the Claude or OpenAI API connected to your own knowledge base via retrieval-augmented generation (RAG), so it answers from real, trusted documents instead of guessing. You'll automate a genuine workflow from your field end to end. The edge it gives you is enormous: 'I built an AI agent that does X for my industry' is the line that gets technical and business grads interviews. Every lesson advances a working build, and you walk away with a deployed, demoable agent plus documentation an employer can poke at.
9 lessons
What an AI Agent Really Is
Cut through the hype: understand the difference between a chatbot, a RAG assistant, and a tool-using agent, and scope a realistic agent for a real workflow in your field. You'll produce an agent spec — purpose, users, knowledge sources, actions.
Calling the Claude / OpenAI API
Make your first real API calls, handle keys safely, and understand requests, responses, tokens, and cost. You'll build a working script that sends a prompt and returns a structured answer — your agent's heartbeat.
Designing the System Prompt and Persona
Engineer the system prompt that defines your agent's role, rules, tone, and refusals so it behaves reliably for real users. You'll deliver a tested system prompt with documented behavior and guardrails.
Building the Knowledge Base (RAG), Part 1
Learn retrieval-augmented generation from the ground up: chunking documents, embeddings, and a vector store so the agent answers from YOUR data. You'll ingest a real document set into a working knowledge base.
RAG Part 2: Grounded, Cited Answers
Wire retrieval into the agent so it answers from the knowledge base and cites its sources — and refuses gracefully when it doesn't know. You'll demo the agent giving grounded, source-backed answers.
Giving the Agent Tools and Actions
Let the agent DO things — call a function, look something up, file an intake record — using tool/function calling. You'll add one real action so your agent automates an actual step, not just talk.
Testing, Guardrails, and Failure Modes
Pressure-test the agent: adversarial inputs, hallucination checks, cost and rate limits, and safe failure. You'll produce a test suite and a guardrails checklist that make the agent trustworthy.
Deploying and Wrapping It in a Usable Interface
Put a simple chat interface on the agent and deploy it so anyone can try it from a link. You'll have a live, shareable agent URL.
Capstone: A Deployed Agent + Demo Walkthrough
Finalize your deployed domain agent and a short demo and write-up — what it does, how RAG grounds it, what workflow it automates. This is the capstone that makes employers say 'tell me more.'