Most people are still prompting. This workshop teaches you to build agents that act.

You'll see a demo of working HR Assistant agent from scratch — live, hands-on, no black boxes. Start with a LlamaIndex RAG pipeline over real HR policy documents. Wrap it as a callable tool. Add a second custom Python tool for live data lookups. Then connect both to a ReAct agent that reasons through Thought → Action → Observation entirely on its own.

By the end, one question triggers the full loop. The agent decides whether to retrieve from documents, call a function, or do both — then fuses the results into a single grounded answer. No hardcoded logic. No guessing. Just an agent that decides.

Key Takeaways

• How to wrap a LlamaIndex RAG pipeline as a callable tool an agent can reason over
• How to build a custom Python tool for live data lookups — and connect it alongside retrieval
• The ReAct pattern in practice: Thought → Action → Observation, running live, not just diagrammed on a slide
• How an agent decides when to retrieve, when to call a function, and when to do both — without hardcoded if/else logic
• A working, extendable HR Assistant agent you can take and adapt to your own use case

Who Is This For

Builders who already know the basics — you've written a prompt, maybe called an API, maybe built a simple RAG pipeline — and want to see how those pieces become an actual agent. This is not an intro to AI session; it assumes you're comfortable with Python and have touched an LLM API before.

If you've been prompting and are ready to build systems that reason and act on their own, this is the next step.

This is the small part from Decoding Data Science's AI Residency - the same pattern powering every serious agentic AI system in production today: retrieval, tool-calling, and reasoning, working as one system instead of three separate tricks.

2 CPD Hours · Live, hands-on build · Part of the AI Residency Masterclass series