SnowPro Specialty – Gen AI (GES-C02)

Learning plan

About 50 hours of focused study, split into 5 phases. Pick your weekly pace and the timeline adjusts.

Estimated finish: 5 weeks

What the exam covers

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

Before you start

• Strong Snowflake SQL and data-engineering basics. SnowPro Core is recommended, and Snowflake suggests 1+ year of GenAI work on Snowflake
• Some Python (Snowpark, Snowflake ML) is helpful
• A free 30-day Snowflake trial account in a region that supports Cortex AI functions

1. Week 1 · 9 h

   Cortex AI landscape, governance and cost

   • Map each Cortex capability to a use case (AISQL/LLM functions, Cortex Search, Cortex Analyst, Cortex Agents, fine-tuning, document AI functions)
   • Explain how Cortex keeps data inside Snowflake's governance boundary, and control model access with roles and the account-level model allowlist
   • Estimate and monitor credit/token cost for LLM functions, and know regional availability and cross-region inference

   Use: Snowflake AI and ML features overview · SnowPro Specialty Gen AI exam page

2. Weeks 1–3 · 12 h

   Cortex AI functions in SQL and Python

   • Call AI_COMPLETE (COMPLETE), AI_CLASSIFY, AI_FILTER, AI_AGG, AI_SENTIMENT, SUMMARIZE, TRANSLATE and EMBED from SQL and from Snowpark Python
   • Pick a model by quality, latency and cost; use structured outputs and inference parameters (temperature, max_tokens)
   • Apply Cortex Guard and prompt patterns to reduce harmful or ungrounded output

   Hands-on: In a trial account, load a few hundred product reviews into a table, then use one SQL statement to classify each by topic, score sentiment, and write a one-line summary per product with AI_AGG. Compare the cost of two models in the usage views.

   Use: Cortex AISQL functions · Snowflake quickstarts

3. Weeks 3–4 · 12 h

   RAG, Cortex Search, Analyst and Agents

   • Create a Cortex Search service over chunked text (target lag, attributes, filters) and query it from SQL and Python
   • Build a RAG flow that feeds Cortex Search results into AI_COMPLETE, and evaluate it with AI Observability (TruLens-based) metrics
   • Write a Cortex Analyst semantic model or view for text-to-SQL, and combine Search and Analyst in a Cortex Agent

   Hands-on: Chunk a set of policy documents into a table, create a Cortex Search service on it, and build a Streamlit-in-Snowflake chat app that answers questions with citations through AI_COMPLETE.

   Use: Cortex Search overview · Cortex Analyst · Cortex Agents · AI Observability

4. Weeks 4–5 · 10 h

   Fine-tuning, custom models and document pipelines

   • Run Cortex Fine-tuning on a training/validation table and call the tuned model; know when fine-tuning beats prompting or RAG
   • Deploy an open-source model from Snowflake Model Registry to Snowpark Container Services (compute pools, services, GPU sizing)
   • Build a document pipeline with AI_PARSE_DOCUMENT (OCR vs layout mode) and AI_EXTRACT over files in a stage, then automate it with streams and tasks

   Hands-on: Upload 10 invoices (PDF) to an internal stage, parse them with AI_PARSE_DOCUMENT in layout mode, pull vendor, date and total with AI_EXTRACT into a table, and schedule the pipeline with a task.

   Use: Cortex Fine-tuning · Snowflake Model Registry · Snowpark Container Services · AI_PARSE_DOCUMENT

5. Week 5 · 7 h

   Practice and review

   • Download the official study guide from the exam page and check every objective against your notes
   • Take a timed practice test and re-study any Cortex function whose arguments or return type you got wrong
   • Review the access-control, cost and regional-availability rules, which are easy points to drop

   Use: Udemy – SnowPro Specialty Gen AI practice tests

You're ready to book when…

• You can write the SQL for any Cortex AI function from memory, with correct arguments
• You have built a Cortex Search RAG app and a document-extraction pipeline in your own trial account
• You can explain when to use fine-tuning, Model Registry with SPCS, or plain prompting for a scenario

Exam tips

• The current exam is GES-C02. Snowflake's C02 page adds document-processing functions and SPCS-based model building, so older GES-C01 material and the retired Document AI content are out of date.
• Snowflake's official study guide is gated behind a short form on the exam page. It is free, and it gives the objective-level detail that the page omits.