watsonx Data Scientist - Associate study plan

Map the watsonx.ai Prompt Lab interface: identify where to select a foundation model, set decoding parameters, and save a prompt asset, then record the exact click path in your notes

Explain in writing the difference between foundation models tuned for generation versus encoder models used for extraction, and name at least three model families available in watsonx.ai with one typical use case each

Experiment with temperature, top_p, and max tokens on the same prompt and document how each parameter shifts output length and variability across five runs

Draft ten prompt variants for a single extraction task and rank them by output quality using a simple rubric you define

Apply few-shot prompting and chain-of-thought patterns to two tasks and compare accuracy against zero-shot baselines

Configure a prompt template with variable placeholders in Prompt Lab and test it against at least three different input values

Describe how tokenization affects cost and context limits, and calculate approximate token counts for a sample document

Identify the guardrails and content-filtering options available when running prompts and note what each blocks

Walk through the model development lifecycle in watsonx.ai from project creation to deployment and sketch it as a diagram

Create a project, associate a runtime instance, and confirm you can open a notebook and run a cell against the environment

Compare tuning approaches (prompt tuning versus full fine-tuning) on cost, data requirements, and expected quality gains

Prepare a small instruction-style training dataset in the required JSONL format and validate its structure

Install and authenticate the watsonx SDKs in a notebook using an API key and project ID, and confirm a successful client call

Write code to list available foundation models and generate text programmatically with specified parameters

Deploy a tuned or prompted model to a deployment endpoint and invoke it from a notebook with a sample payload

Handle common SDK errors such as authentication failures and token limits by adding basic retry and error handling

Run an AutoAI experiment on a tabular dataset and interpret the pipeline leaderboard, including how preprocessing steps are ranked

Explain the difference between AutoAI and manual pipeline construction and identify when each is appropriate

Inspect a generated pipeline, identify the winning estimator, and note which hyperparameters were tuned

Save and deploy the best AutoAI pipeline, then score new records against it

Build a retrieval-augmented generation pipeline that chunks documents, embeds them, stores vectors, and retrieves relevant passages for a query

Compare chunking strategies and embedding choices, and document retrieval quality on at least five test questions

Connect watsonx.data as a data source and query it from a notebook or the console to confirm connectivity

Explain how watsonx.governance tracks model factsheets and risk metadata across the lifecycle

Define evaluation metrics for both generative and predictive models and apply at least three to a sample model

Set up a model evaluation run in watsonx.governance and interpret the fairness and drift outputs

Identify bias detection metrics and describe when each is appropriate for a classification model

Document how factsheets, model cards, and approval workflows support responsible AI governance

Diagram the watsonx platform architecture showing how watsonx.ai, watsonx.data, and watsonx.governance interact with shared services

Complete a timed 60-question practice set across all eight domains and log weak areas by domain

Re-review the two lowest-scoring domains using your own notes and rewrite the weakest concept in plain language

Build a one-page exam-day reference of key terms, SDK calls, and governance concepts