My Skill Lab
Artificial IntelligenceAdvanced

Building AI Agents & RAG Systems

Design agents that use tools, memory and your own data.

Taught by Asha Karki · AI Product Engineer

What you'll learn

  • Define tools as JSON schemas and run a correct function-calling loop, including parallel calls, error handling, and result feedback
  • Build a planning agent with a ReAct-style reason-act-observe loop, a scratchpad, and a stop condition that avoids infinite loops
  • Design short-term and long-term memory: conversation buffers, summarization, and semantic memory backed by a vector store
  • Construct a RAG pipeline end to end: chunk documents, generate embeddings, run vector search, and rerank for relevance
  • Evaluate agents and RAG with retrieval and answer metrics, add guardrails for safety and grounding, and trap failures
  • Control cost and latency with model routing, caching, token budgeting, and concurrency so the system is affordable in production

About this course

Go beyond chat completions and build LLM systems that take action: agents that call your APIs, plan multi-step work, remember context, and answer from your own documents instead of hallucinating. You will learn the full stack of production agents and retrieval-augmented generation (RAG) from tool schemas and planning loops to chunking, embeddings, vector search, and reranking. By the end you can ship an agent that is grounded, evaluated, guard-railed, and cost-aware.

Course content

8 lessons
  1. 1What an LLM Agent Actually IsMove from one-shot prompts to agents that observe, decide, and act in a loop.Preview
  2. 2Tool Use and Function CallingDefine tools as schemas, parse the model's calls, and feed results back safely.Preview
  3. 3Planning Loops: ReAct and BeyondGive the agent a way to reason, act, observe, and decide when it is done.
  4. 4Memory: Short-Term and Long-TermManage the context window and persist what matters across sessions.
  5. 5RAG Part 1: Chunking and EmbeddingsTurn documents into searchable vectors without shredding their meaning.
  6. 6RAG Part 2: Vector Search and RerankingRetrieve candidates fast, then rerank for precision before prompting.
  7. 7Grounding, Evaluation, and GuardrailsMeasure whether the agent is right and stop it from going wrong.
  8. 8Cost, Latency, and ProductionMake the agent fast and affordable, then ship and monitor it.