Study plan - SnowPro Gen AI C02

Founder profile

1+ year hands-on Snowflake Gen AI (meets prereq)
Strong Python + SQL/DE
Employed at phData (Snowflake Premier Partner) - access to internal channels + likely seen most concepts in client work
Cost: $375 + $5k base bump on pass = trivially positive EV

This is not a "study from zero" plan. It's a gap-find-and-close plan with two tracks:

Pick a track

Track A: Sprint (7-10 days, 1-2 hr/day)

For if founder wants to take the exam within 2 weeks of launch (early enough to be a phData first-mover, plus the practical knowledge is already there).

Track B: Steady (3-4 weeks, 30-60 min/day)

For if founder wants to fold this in around other work without rushing.

Recommendation: Track A. Founder's prior experience + DE/Python skills + risk profile (single $375 attempt cost, immediate $5k upside, plus phData first-mover positioning for the bonus program) all argue for sprinting.

Reading order (Track A, by day)

Day	Focus	Reading	Hands-on
Day 1	Set up + scope	README + study-guide-summary.md. Log in to Snowflake learning portal, download official C02 study guide PDF. Register for exam.	Spin up Snowflake trial account (or use phData sandbox). Confirm `SNOWFLAKE.CORTEX_USER` role granted.
Day 2	Cortex AI functions, part 1	study-cortex-ai.md sections: namespaces, AI_* functions, legacy SNOWFLAKE.CORTEX.*	Run COMPLETE, EMBED_TEXT_*, SUMMARIZE, TRANSLATE, SENTIMENT against sample data. Compare AI_COMPLETE vs COMPLETE outputs.
Day 3	Cortex AI functions, part 2	study-cortex-ai.md sections: Cortex Search, Cortex Analyst, fine-tuning, RAG pipeline	Build a tiny RAG end-to-end: stage 3 PDFs → PARSE_DOCUMENT → CORTEX SEARCH SERVICE → query with AI_COMPLETE. This is the single highest-value exercise.
Day 4	Document processing	study-document-processing.md	Test PARSE_DOCUMENT in both OCR and LAYOUT modes. Inspect the JSON output. Compare with AI_EXTRACT given a schema.
Day 5	SPCS + Model Registry (now Domain 1.0 overview-level, ~part of 18%)	study-snowpark-container-services.md + study-model-registry.md — read at concept level, do NOT over-invest	Push a sklearn model to Model Registry, log a version, run inference from SQL. SPCS hands-on optional. Per official C02, this is overview material, not its own domain.
Day 6	Governance — Domain 3.0 (29%, 2nd-largest domain)	`study-governance.md` (3.1 model access · 3.2 RBAC · 3.3 cost · 3.4 observability) + Sharma blog post.	Query `SNOWFLAKE.ACCOUNT_USAGE.CORTEX_FUNCTIONS_QUERY_HISTORY`. Set/inspect `CORTEX_MODELS_ALLOWLIST`. Walk RBAC for who-can-call-which-Cortex-function. This is a full domain, not a review.
Day 7	Practice + gap close	flashcards-deck.md (run through twice). Snowflake University practice questions if available.	Take any practice exam you trust. Note wrong answers. Re-read the corresponding study file.
Day 8-9 (buffer)	Re-read weakest areas + repeat flashcards		Run through the RAG pipeline once more from memory without copy-paste.
Day 10	Exam day		Light review of flashcards morning-of; full night sleep.

Spend-extra-time topics (per community signal)

These are commonly under-prepared given how heavily they're tested:

RBAC + permissions — which roles can call which functions, CORTEX_USER database role, CORTEX_MODELS_ALLOWLIST account parameter, service ownership in SPCS
Cost model nuance — token billing vs warehouse-credit billing vs Cortex Search multi-component billing. Don't memorize prices, memorize which component drives cost.
Function name discrimination — AI_EXTRACT vs PARSE_DOCUMENT vs EXTRACT_ANSWER are easy to conflate. AI_AGG vs SUMMARIZE_AGG. AI_COMPLETE vs COMPLETE (latter is legacy, both still tested).
Embedding model dimensions — 768 vs 1024 mismatch is a trap question
When to use Model Registry SPCS deployment vs warehouse deployment — GPU is the answer
Cross-region inference — when it kicks in + governance implications

Hands-on labs (the irreplaceable part)

Cortex Playground (in Snowsight) lets you experiment with temperature, top_p, and safety settings interactively. Spend ~30 min here on Day 2. Then run all the SQL examples in study-cortex-ai.md against live data on Days 2-4. Founder already has Snowflake access via phData; use a personal sandbox or the trial account to avoid touching client data.

Question format

~55 questions (per C01; C02 likely same)
85 minutes (~93 sec / question)
Multiple-choice + multiple-select + some interactive scenario questions
Scenario framing: "You need to build X for Y. Which Snowflake feature do you pick?" Read all options before answering — Snowflake exams love distractors that would work but aren't best-practice.
Scaled passing score: 750/1000

What to do morning of exam

Run through flashcards once
Skim study-cortex-ai.md function tables one more time (they're the densest material)
Do NOT try to learn anything new; trust the prep
Check date for time zone of the exam slot, log in 15 min early

Post-exam

Whether pass or fail: write a post-mortem-c02-attempt-N.md in this folder
- Topics that surprised you
- Topics overcovered
- Resources that turned out to be high-signal vs low-signal
- Update this study-plan.md for the next person (or your next cert)
On pass: file the cert PDF in ~/rdco-vault/03-personal/certifications/
On pass: trigger phData certification bonus claim via internal HR process

Reusable template note

This plan structure (README + per-domain study notes + plan + flashcards + third-party + post-mortem) is designed to clone for the next Snowflake cert. After this exam, promote the skeleton to ~/rdco-vault/01-projects/certifications/_template/ and reuse.