CNLP

Those figures are set by the certifying organization and change between delivery windows, so check its official website for the current question count and time limit. Do not plan around numbers from older candidate reports. What matters for pacing is practicing under a fixed clock so you learn how long you can spend on a multi-part scenario.

The blueprint implies a computer-delivered, multiple-choice exam with scenario-based items that combine a short situation with a decision. Some questions ask you to pick the best approach rather than the only correct fact. Confirm the exact delivery format, including whether any items are multi-select, on the official site before booking.

It is a working-practitioner exam, not an entry-level vocabulary test. The hard parts are tradeoff questions: choosing between a rule-based pipeline and a fine-tuned transformer, or spotting why an evaluation metric misleads in production. Candidates who have only read tutorials tend to struggle with deployment, monitoring, and ethics items.

Retake rules, waiting periods, and any fees are set by the certifying body and are not fixed across providers. Look up the current retake policy on its official site before you sit the exam. If a retake is allowed, use your score report to identify weak domains rather than restudying everything evenly.

Start with fundamentals and preprocessing, because tokenization and normalization errors propagate into every later topic. Then cover classical methods and deep learning, which build the vocabulary for transformers. Finish with evaluation, ethics, deployment, and use cases, since those questions assume you already understand the modeling choices being evaluated.

Three patterns recur: confusing evaluation metrics, such as treating perplexity as a measure of task quality; ignoring data leakage between train and test splits during preprocessing; and answering deployment questions with modeling solutions. Read each scenario for the constraint being tested, such as latency, label scarcity, or auditability, before choosing an option.

You should understand attention conceptually, why positional information is needed, how tokenization affects sequence length and cost, and the difference between pretraining and fine-tuning. You are unlikely to be asked to derive gradients. Focus on the consequences of architectural choices for accuracy, latency, and memory.

The exam is knowledge-based, but hands-on practice is the most efficient way to learn. Build at least one end-to-end pipeline: preprocess text, train a baseline and a transformer, evaluate both, and deploy one behind an API with logging. That experience makes scenario questions about monitoring, drift, and failure modes far easier.

This page is independent study material and is not affiliated with, endorsed by, or sponsored by the certifying organization. Always treat the official body's published handbook as the authoritative source for eligibility, logistics, and scoring, and use independent guides only to structure your preparation.