Learn. Build. Advance.

Start with a concrete outcome

Use the sequences below to choose what to read and what to produce. The links lead to published guides; the suggested tasks are exercises for you to complete, not claims of measured results.

1. Diagnose a RAG system

Prerequisites: basic Python and an understanding of retrieval and embeddings.

1. Read why RAG returns wrong answers and list the failure modes relevant to your documents.
2. Work through the 30-question evaluation exercise. Save the fictional corpus, record retrieved source IDs, and compare your answers with the answer key.
3. Compare chunking strategies using the same questions. Change one setting at a time.

Deliverable: an evaluation log showing a baseline, one change, and the failures that remain.

2. Add controls to an LLM application

Prerequisites: Python, environment variables, and a model-provider account if you run hosted examples. Check each guide’s setup requirements and provider costs before running code.

1. Compare Llama Guard and NeMo Guardrails to separate classification, orchestration, and application permissions.
2. Follow the NeMo Guardrails and LangChain guide and document the controls you configure.
3. Use the LLM-as-a-judge guide to plan evaluation alongside human-reviewed examples.

Deliverable: a small test matrix covering normal requests, blocked requests, malformed responses, and provider failures. Label mocked tests separately from live model tests.

3. Prepare for interviews and certification

Begin with the LLM system design interview framework. Explain your design choices using an example from your own work, then use the independent practice tests to identify topics to revisit. Check the certifying body’s current official exam guide before scheduling an exam.

Deliverable: a design walkthrough and a list of topics you missed, with an explanation of each correction.