IBM Certified watsonx Generative AI Engineer – Associate (C1000-185)

Learning plan

About 40 hours of focused study, split into 4 phases. Pick your weekly pace and the timeline adjusts.

Estimated finish: 4 weeks

What the exam covers

Source: official exam guide. Weights can change, so check before you start.

Before you start

• Python and basic data-science skills (IBM lists Python, data analysis and data science)
• An IBM Cloud account with a watsonx.ai trial or Lite plan for Prompt Lab, Tuning Studio and deployments
• IBM assumes 6–12 months of hands-on watsonx experience, so plan extra lab time if you are new to it

1. Week 1 · 10 h

   Solution design and prompt engineering

   • Match a use case to a model family on watsonx.ai (Granite, Llama, Mistral and others) by task, context length, licence and cost
   • Explain LLM limits (hallucination, context limits, bias) and when GenAI is the wrong tool
   • Use Prompt Lab in chat, structured and freeform modes, and tune decoding (greedy vs sampling, temperature, top-k/top-p, repetition penalty, stop sequences)
   • Save and reuse prompt templates with variables, and evaluate prompts with watsonx.governance metrics

   Hands-on: In watsonx.ai Prompt Lab, build an extraction prompt for support emails. Compare greedy and sampling decoding on 10 examples, save it as a prompt template with a variable, and export the Python code.

   Use: Exam objectives (IBM certification page) · Prompt Lab documentation · IBM learning path for this certification

2. Weeks 2–3 · 12 h

   Fine-tuning (largest domain, 31%)

   • Choose between prompt engineering, prompt tuning and fine-tuning, and explain parameter-efficient methods (LoRA, prompt tuning)
   • Prepare training data in the JSONL input/output format, set epochs, learning rate and batch size in Tuning Studio, and read the loss curve
   • Explain InstructLab's taxonomy-driven synthetic data generation and how it differs from classic fine-tuning
   • Deploy and test a tuned model, then compare it against the base model

   Hands-on: Prepare 100+ labelled JSONL examples for a classification task, run a tuning experiment in Tuning Studio, deploy the tuned model, and compare its accuracy with the zero-shot base model on 20 held-out examples.

   Use: Tuning Studio documentation · InstructLab (GitHub)

3. Weeks 3–4 · 12 h

   RAG, deployment and orchestration

   • Build RAG on watsonx.ai with chunking, an embedding model, a vector store (Milvus or Elasticsearch) and grounded prompts
   • Deploy prompt templates and tuned models to deployment spaces, and call them through the REST API and the ibm-watsonx-ai Python SDK
   • Integrate watsonx.ai models with LangChain/LlamaIndex and agent frameworks for multi-step orchestration

   Use: RAG on watsonx.ai documentation · ibm-watsonx-ai Python SDK · IBM watsonx.ai samples (GitHub)

4. Week 4 · 6 h

   Exam prep and review

   • Watch IBM's certification prep session and map each objective to your lab notes
   • Re-drill Fine-Tuning (31%), then RAG (17%) and Prompt Engineering (16%)
   • Answer 62 timed practice questions in 90 minutes and aim for at least 44 correct

   Use: IBM Certification Prep Session (Mediacenter)

You're ready to book when…

• You have tuned, deployed and queried a model in watsonx.ai yourself
• You can explain prompt tuning vs LoRA fine-tuning vs InstructLab and pick one for a scenario
• You can build a RAG notebook with the Python SDK without copying a sample
• You get at least 44 of 62 questions right on a timed practice set

Exam tips

• Exam C1000-185 has 62 questions and you need 44 correct to pass (about 71%), with 90 minutes allowed. IBM's page lists the price as $200 USD.
• Fine-Tuning is almost a third of the exam, which is unusual for an associate exam. Do the Tuning Studio lab rather than only reading about it.