Mocka logoMocka
Home
Why MockaPricingFAQAbout

Google Cloud Certified Generative AI Leader Ultimate Cheat Sheet

4 Domains • 30 Concepts • Approx. 4 pages

Your Quick Reference Study Guide

This cheat sheet covers the core concepts, terms, and definitions you need to know for the Google Cloud Certified Generative AI Leader. We've distilled the most important domains, topics, and critical details to help your exam preparation.

💡 Note: While this study guide highlights essential concepts, it's designed to complement—not replace—comprehensiv e learning materials. Use it for quick reviews, last-minute prep, or to identify areas that need deeper study before your exam.

Google Cloud Certified Generative AI Leader Practice Questions
Access Mock Exams & Comprehensive Question Bank
Listen to Audio Podcasts
Expert summaries for Google Cloud Certified Generative AI Leader

About This Cheat Sheet: This study guide covers core concepts for Google Cloud Certified Generative AI Leader. It highlights key terms, definitions, common mistakes, and frequently confused topics to support your exam preparation.

Use this as a quick reference alongside comprehensive study materials.

Google Cloud Certified Generative AI Leader

Cheat Sheet •

Provided by GetMocka.com

About This Cheat Sheet: This study guide covers core concepts for Google Cloud Certified Generative AI Leader. It highlights key terms, definitions, common mistakes, and frequently confused topics to support your exam preparation.

Use this as a quick reference alongside comprehensive study materials.

Fundamentals of Generative AI

30%

Unsupervised Learning — Clusters, Embeddings, Anomalies

Learn structure from unlabeled data—clustering, dimensionality reduction, embeddings, anomaly detection for downstreams.

Key Insight

Produces representations/groups, not ground-truth labels — validate with metrics or downstream tasks to avoid noise.

Often Confused With

ClusteringSemi-supervised Learning

Common Mistakes

  • Equating unsupervised learning to only clustering.
  • Skipping validation because data has no labels.
  • Treating every discovered pattern as business-meaningful without testing.

Diffusion Models — Iterative Denoising Generators

Generative models that iteratively denoise random noise into data; strong for high-quality image and multimodal outputs.

Key Insight

They learn to reverse a noise process via many stochastic denoising steps — sampling is iterative (and usually stochastic) unless accelerated.

Often Confused With

GANsAutoregressive Models

Common Mistakes

  • Assuming diffusion models are the same as GANs (they're different training/objectives).
  • Expecting single-feedforward sampling — standard sampling is iterative.
  • Believing SOTA quality needs little data/compute — top results usually need large models or tuning.

Structured vs Unstructured Data — OCR, Chunking & Embeddings

Tabular vs text/image/audio/video; use OCR, chunking and embeddings to index and retrieve unstructured data for gen AI.

Key Insight

Unstructured becomes usable after preprocessing: OCR → chunking → embeddings; pick vector store vs text index by latency, cost, and compliance.

Often Confused With

Data qualityVector embeddingsRelational databases

Common Mistakes

  • Cloud storage ≠ instantly usable — you still need preprocessing and indexing.
  • Believing unstructured can't be searched — embeddings/text indexes enable retrieval.
  • Assuming more raw unstructured data always improves models; relevance and label quality matter more.

Data Quality — Accuracy, Labels & Drift

Accuracy, completeness, timeliness, consistency and label quality — the fitness signals that determine model reliability

Key Insight

Label quality and freshness often beat brute volume; continuously monitor label noise, data drift, and business KPI impact.

Often Confused With

Data lineageModel evaluationLabeling

Common Mistakes

  • Assuming more volume always improves models — noisy data degrades performance.
  • Treating cleaning as one-off — skip ongoing monitoring and drift detection at your peril.
  • Believing data quality only matters for training, not inference or business outcomes.

Model Hallucinations — Factuality Failures

When a model fabricates unsupported facts; detect with grounding, tool calls, verification, and monitoring.

Key Insight

No single cure: combine grounding (RAG/tool calls), constrained decoding, post‑answer verification, and monitoring.

Often Confused With

RAG (Retrieval‑Augmented Generation)Model BiasCalibration/Confidence

Common Mistakes

  • Thinking prompt tweaks or temperature settings eliminate hallucinations
  • Assuming large commercial models never hallucinate
  • Believing RAG or provenance completely removes unsupported claims

Prompt Engineering: Templates, Context & Controls

Craft, version, test and guard prompts/templates to steer LLMs safely, reproducibly, and locally.

Key Insight

Prompts shape behavior but don't create facts—use templates, tests, localization, versioning and pair with grounding.

Often Confused With

Fine‑tuningRAG (Retrieval‑Augmented Generation)Decoding Parameters

Common Mistakes

  • Relying on prompt wording alone for factual accuracy
  • Assuming longer or cleverer prompts always improve outputs
  • Expecting one prompt to work for all languages and contexts

RAG — Prebuilt vs Custom (Vertex AI Search)

Ground LLMs by retrieving enterprise docs and invoking tools to reduce hallucinations and produce auditable answers.

Key Insight

Prebuilt RAG = faster, less ops; custom RAG = control over embeddings, vector store, relevance and compliance. Retrieval supplies context; tool calls*

Often Confused With

Fine-tuningTool/API invocationVector stores & embeddings

Common Mistakes

  • Assuming managed RAG removes the need for data governance or security controls
  • Treating tool invocation (actions) as identical to returning retrieved text
  • Believing more retrieved docs always improves answer accuracy

AI Data Protection: DPIA, Controls & Risks

Technical, operational, and contractual controls (DPIA/AIA, anonymization limits, encryption, residency, retention) to护护

Key Insight

Consent, contractual clauses, or private deployments don't remove risk—run an AIA/DPIA, map data flows, and apply technical+contractual controls sized

Often Confused With

User consentAnonymizationPrivate deployment/VPC

Common Mistakes

  • Relying on user consent alone to meet data-protection obligations
  • Assuming simple removal of identifiers fully eliminates privacy risk for model tuning
  • Believing a vendor contract guarantees compliance regardless of how or where data is processed

Unsupervised Learning — Clusters, Embeddings, Anomalies

Learn structure from unlabeled data—clustering, dimensionality reduction, embeddings, anomaly detection for downstreams.

Key Insight

Produces representations/groups, not ground-truth labels — validate with metrics or downstream tasks to avoid noise.

Often Confused With

ClusteringSemi-supervised Learning

Common Mistakes

  • Equating unsupervised learning to only clustering.
  • Skipping validation because data has no labels.
  • Treating every discovered pattern as business-meaningful without testing.

Diffusion Models — Iterative Denoising Generators

Generative models that iteratively denoise random noise into data; strong for high-quality image and multimodal outputs.

Key Insight

They learn to reverse a noise process via many stochastic denoising steps — sampling is iterative (and usually stochastic) unless accelerated.

Often Confused With

GANsAutoregressive Models

Common Mistakes

  • Assuming diffusion models are the same as GANs (they're different training/objectives).
  • Expecting single-feedforward sampling — standard sampling is iterative.
  • Believing SOTA quality needs little data/compute — top results usually need large models or tuning.

Structured vs Unstructured Data — OCR, Chunking & Embeddings

Tabular vs text/image/audio/video; use OCR, chunking and embeddings to index and retrieve unstructured data for gen AI.

Key Insight

Unstructured becomes usable after preprocessing: OCR → chunking → embeddings; pick vector store vs text index by latency, cost, and compliance.

Often Confused With

Data qualityVector embeddingsRelational databases

Common Mistakes

  • Cloud storage ≠ instantly usable — you still need preprocessing and indexing.
  • Believing unstructured can't be searched — embeddings/text indexes enable retrieval.
  • Assuming more raw unstructured data always improves models; relevance and label quality matter more.

Data Quality — Accuracy, Labels & Drift

Accuracy, completeness, timeliness, consistency and label quality — the fitness signals that determine model reliability

Key Insight

Label quality and freshness often beat brute volume; continuously monitor label noise, data drift, and business KPI impact.

Often Confused With

Data lineageModel evaluationLabeling

Common Mistakes

  • Assuming more volume always improves models — noisy data degrades performance.
  • Treating cleaning as one-off — skip ongoing monitoring and drift detection at your peril.
  • Believing data quality only matters for training, not inference or business outcomes.

Model Hallucinations — Factuality Failures

When a model fabricates unsupported facts; detect with grounding, tool calls, verification, and monitoring.

Key Insight

No single cure: combine grounding (RAG/tool calls), constrained decoding, post‑answer verification, and monitoring.

Often Confused With

RAG (Retrieval‑Augmented Generation)Model BiasCalibration/Confidence

Common Mistakes

  • Thinking prompt tweaks or temperature settings eliminate hallucinations
  • Assuming large commercial models never hallucinate
  • Believing RAG or provenance completely removes unsupported claims

Prompt Engineering: Templates, Context & Controls

Craft, version, test and guard prompts/templates to steer LLMs safely, reproducibly, and locally.

Key Insight

Prompts shape behavior but don't create facts—use templates, tests, localization, versioning and pair with grounding.

Often Confused With

Fine‑tuningRAG (Retrieval‑Augmented Generation)Decoding Parameters

Common Mistakes

  • Relying on prompt wording alone for factual accuracy
  • Assuming longer or cleverer prompts always improve outputs
  • Expecting one prompt to work for all languages and contexts

RAG — Prebuilt vs Custom (Vertex AI Search)

Ground LLMs by retrieving enterprise docs and invoking tools to reduce hallucinations and produce auditable answers.

Key Insight

Prebuilt RAG = faster, less ops; custom RAG = control over embeddings, vector store, relevance and compliance. Retrieval supplies context; tool calls*

Often Confused With

Fine-tuningTool/API invocationVector stores & embeddings

Common Mistakes

  • Assuming managed RAG removes the need for data governance or security controls
  • Treating tool invocation (actions) as identical to returning retrieved text
  • Believing more retrieved docs always improves answer accuracy

AI Data Protection: DPIA, Controls & Risks

Technical, operational, and contractual controls (DPIA/AIA, anonymization limits, encryption, residency, retention) to护护

Key Insight

Consent, contractual clauses, or private deployments don't remove risk—run an AIA/DPIA, map data flows, and apply technical+contractual controls sized

Often Confused With

User consentAnonymizationPrivate deployment/VPC

Common Mistakes

  • Relying on user consent alone to meet data-protection obligations
  • Assuming simple removal of identifiers fully eliminates privacy risk for model tuning
  • Believing a vendor contract guarantees compliance regardless of how or where data is processed

Google Cloud’s Generative AI Offerings

35%

Model Management & Governance (Registry, Lineage, Audit)

Platform lifecycle + provenance: registry, CI/CD gates, monitoring and audit logs for compliant production use.

Key Insight

Governance is cross‑functional — enforce with registry metadata, CI/CD gating, versioned lineage, and audit logs.

Often Confused With

MLOpsData GovernanceModel Monitoring

Common Mistakes

  • Thinking governance is only a legal/compliance task; it's cross‑functional.
  • Enabling automatic model upgrades without validation, testing, or rollback plans.
  • Treating a model registry as mere file storage instead of metadata/version/lineage store.

Fine-tuning & PEFT (LoRA, Adapters, Instruction Tuning)

Specialize foundation models via full, instruction, or parameter‑efficient tuning; weigh data, cost, latency, and risk.

Key Insight

Match tuning method to constraints: PEFT for small/private data and low cost/latency; full fine‑tune when major domain shift requires it.

Often Confused With

Prompt EngineeringRAG (Retrieval-Augmented Generation)Transfer Learning

Common Mistakes

  • Assuming fine‑tuning always means retraining all model weights; PEFT alters far fewer params.
  • Believing fine‑tuning always outperforms prompt engineering or RAG on cost, latency, and safety.
  • Expecting massive labeled datasets are always required; small, high‑quality sets often suffice.

Secure AI & SAIF — Controls Mapped to ML Lifecycle

Map IAM, network, KMS, logging and monitoring to ML lifecycle stages to manage risk and auditability.

Key Insight

Layer controls by lifecycle stage — least privilege IAM + network boundaries + KMS + audit logging; vendor certs don't remove customer duties.

Often Confused With

RBAC/ABACVPC Service ControlsShared Responsibility Model

Common Mistakes

  • Thinking encryption at rest removes the need for strict access controls and logging.
  • Assuming RBAC alone provides context-aware authorization; skipping ABAC/conditional rules.
  • Treating vendor compliance certificates as automatic customer compliance evidence.

Gemini Enterprise — NotebookLM, Multimodal Search & Agents

Document-centric enterprise bundle: ingest, index, ground sources for long-context QA, summarization, and agents.

Key Insight

NotebookLM API is for source-grounded doc workflows — it needs indexing, chunking, metadata, and human verification for reliable outputs.

Often Confused With

Public Gemini modelGeneral-purpose LLMsNotebookLM consumer UI

Common Mistakes

  • Assuming NotebookLM API always produces accurate answers and citations without human review.
  • Skipping retrieval prep — believing no indexing, chunking, or metadata is required.
  • Treating Gemini Enterprise as just a larger public Gemini model with no added integrations or controls.

CX Suite — Conversational Agents, Agent Assist, Insights & CCaaS

Integrated gen‑AI CX stack: virtual agents, real‑time agent augmentation, analytics, and cloud contact‑center ops.

Key Insight

Use virtual agents for high‑volume, scripted queries; Agent Assist for live agent augmentation and compliance; Insights closes the loop via analytics—

Often Confused With

Standalone chatbotsIVR systemsOn‑prem contact center

Common Mistakes

  • Assuming conversational agents fully replace human agents in every scenario
  • Expecting Agent Assist to auto‑resolve issues without human oversight or escalation rules
  • Treating CCaaS the same as on‑prem contact centers (ignores elastic scaling, updates, and ops model)

Google Search Grounding for RAG (Public Web)

Ground LLM outputs with live web sources via Google Search APIs—good for freshness and citations, but not a substitute‑P

Key Insight

Public search gives freshness and citations but still has licensing, privacy, paywall, and SLA limits—use for public facts or as supplement to private

Often Confused With

Enterprise site searchWeb scrapingPrivate RAG

Common Mistakes

  • Assuming public search removes copyright or downstream licensing obligations
  • Relying on public search to access paywalled or intranet content
  • Treating public Google Search as unlimited or enterprise‑grade SLA and compliance

Vertex AI Platform — Model Garden & AutoML

Unified GCP platform for Model Garden, AutoML, custom training, hosting, forecasting, and MLOps for production models.

Key Insight

Vertex unifies auto and custom workflows — AutoML speeds prototyping, but production needs data prep, evaluation, hosting, and MLOps pipelines.

Often Confused With

AutoMLCustom trainingModel hosting

Common Mistakes

  • Assuming AutoML yields production‑ready models without data cleaning, evaluation, or MLOps
  • Believing Vertex only runs AutoML and cannot host or custom‑train user models
  • Thinking fine‑tuning capabilities and costs are identical across all foundation models

Vertex AI Fine‑tuning (LoRA & Full FT)

Console, SDK, and APIs to fine‑tune Gemini and supported open models via full parameter tuning or LoRA, with pipelines/监

Key Insight

Fine‑tuning can materially improve domain accuracy, but quality vs cost/latency tradeoffs matter — LoRA lowers cost but may not match full FT.

Often Confused With

Prompt engineeringFull‑parameter fine‑tuning

Common Mistakes

  • Assuming fine‑tuning always beats prompt engineering for every task
  • Believing Vertex fine‑tuning only supports Gemini and can't tune open models
  • Expecting LoRA to match full‑parameter FT quality and have identical latency/cost

Agent Tools: Extensions, Functions, Plugins, Data Stores

Agent interfaces to extend capability—execute code, access grounded data, or integrate third‑party services.

Key Insight

Functions run logic/APIs; data stores only provide grounded content; plugins are prebuilt external integrations with their own auth/lifecycle.

Often Confused With

PluginsFunctionsExtensions

Common Mistakes

  • Assuming plugins and functions are interchangeable (they differ in lifecycle and auth).
  • Believing data stores execute business logic — they only store and serve grounded content.
  • Thinking extensions always require code changes — some are config/declarative.

Grounding Sources: Enterprise Data vs Google Search (RAG)

Choose and prep enterprise stores (Cloud Storage, BigQuery, Firestore) for RAG; Google Search can supplement but is less

Key Insight

Enterprise stores give control, traceability, and auditability but need ingestion, preprocessing, indexing and refresh; Search is fast but volatile/un

Often Confused With

Google SearchCloud StorageBigQuery

Common Mistakes

  • Assuming grounding eliminates hallucinations — it reduces but doesn't remove them.
  • Treating Google Search as authoritative for business decisions.
  • Index once and forget — embeddings/indexes need periodic refresh and reingestion.

Techniques to Improve Generative AI Model Output

20%

Foundation Model Limits & Fixes

Data-dependent limits (cutoff, bias, hallucinations, edge cases); mitigate with RAG, PEFT, prompt design, and HITL.

Key Insight

Match mitigation to the failure: RAG for missing facts, PEFT for domain adaptation, prompts for intent/format, HITL for safety/edge cases — model 'lik

Often Confused With

Prompt engineeringFine‑tuning/PEFTRetrieval‑augmented generation (RAG)

Common Mistakes

  • Assuming human-in-the-loop (HITL) scales linearly or is a universal cure.
  • Trusting model confidence scores as proof of factual correctness.
  • Relying on prompt tweaks alone to eliminate hallucinations.

Bias & Fairness: Metrics to Act On

Detect and reduce disparate impact via subgroup and counterfactual tests, targeted metrics, and defined remediation.

Key Insight

Fairness is a trade-off: pick appropriate metrics (e.g., demographic parity, equalized odds, calibration), run subgroup + counterfactual audits, set S

Often Confused With

Model calibrationData rebalancingExplainability

Common Mistakes

  • Believing more data or simple rebalancing will always remove bias.
  • Assuming improving a metric for one group fixes overall fairness.
  • Thinking explainability alone resolves fairness harms.

Chain-of-Thought (CoT) — Stepwise Reasoning

Prompt the model to show intermediate steps before the final answer to improve multi-step reasoning.

Key Insight

Helps multi-step problems on larger models but often increases verbosity, token cost, latency, and hallucination risk.

Often Confused With

Few-shot promptingZero-shot prompting

Common Mistakes

  • Assuming CoT eliminates hallucinations — it can still produce plausible but wrong steps.
  • Believing CoT requires model fine-tuning; it's applied via prompting.
  • Assuming CoT always improves accuracy; benefits depend on model size and task.

Few-Shot Prompting — In-Context Examples

Put 2–8 concise input→output examples in the prompt so the model imitates the task without changing weights.

Key Insight

Example quality, order, and formatting shape behavior; more examples can harm performance or exceed context limits.

Often Confused With

Fine-tuningFew-shot learning (parameter-based)

Common Mistakes

  • Confusing few-shot prompting with fine-tuning — examples don't update model parameters.
  • Adding many examples always helps — too many or low-quality examples can degrade output or hit context limits.
  • Ignoring example order and formatting — sequence, phrasing, and punctuation affect results.

Grounding Sources — 1st‑party, 3rd‑party, World Data

Constrain outputs with retrievable, governed sources (enterprise, licensed, or web/knowledge graphs) to cite evidence.

Key Insight

Grounding works only with curated, searchable sources + citation; uncurated or missing retrieval = more hallucinations.

Often Confused With

Retrieval‑Augmented Generation (RAG)Knowledge GraphsLive web / Web crawling

Common Mistakes

  • Assuming grounding guarantees facts/up‑to‑date answers regardless of retrieval quality
  • Treating Google's Knowledge Graph as a private enterprise data store
  • Skipping privacy, licensing, or security review for public or licensed web data

Sampling & Safety Controls (temp, top‑p, tokens)

Adjust temperature, top‑p/top‑k, length, stop sequences and safety filters to trade creativity, determinism, cost, and风险

Key Insight

Temperature and top‑p affect randomness differently; long outputs, big beams, or lax safety amplify errors and cost—tune to metric.

Often Confused With

Beam SearchGreedy/Deterministic DecodingSafety Filters / Moderation

Common Mistakes

  • Using top‑p and temperature interchangeably instead of tuning both
  • Assuming longer outputs always reduce hallucinations; errors can compound
  • Relying on stop sequences to perfectly truncate output in every case

Business Strategies for a Successful Generative AI Solution

15%

Foundation Models — Modality, Size & Adaptation

Pretrained text/image/multimodal models; pick modality, size, hosting and adaptation to balance capability, cost, latenc

Key Insight

Match modality & size to the use case; use PEFT/LoRA or prompt tuning to cut cost/time while keeping domain performance

Often Confused With

Large Language Models (LLMs)Fine‑tuned modelsManaged vs Self‑hosted models

Common Mistakes

  • Assuming bigger = always better — ignore cost, latency, and diminishing returns
  • Treating all models as equal for privacy/governance — vendor and hosting choices change risks
  • Believing out‑of‑the‑box models need no adaptation to reach production accuracy

Vertex AI Search — Managed RAG & Semantic Retrieval

Managed RAG: index enterprise content, semantic retrieval, connectors, and conversational grounding to ground LLMs

Key Insight

It's a retrieval + indexing layer to ground LLMs — you still choose the model, tune retrieval relevance, and craft prompts

Often Confused With

Google Cloud SearchVector DBsVertex AI Matching Engine

Common Mistakes

  • Thinking Vertex AI Search is an LLM — it's retrieval/indexing, not the generative model
  • Assuming retrieval needs no tuning — retrieval configuration and prompt design still drive accuracy
  • Expecting zero hallucinations — poor retrieval or prompts still yield incorrect answers

IAM (Identity & Access Management) — Least‑Privilege Gatekeeper

Role-based access control for users, groups, and service accounts to lock down datasets, models, and AI workloads.

Key Insight

Roles + bindings + conditions implement least privilege — service accounts and resource-level policies matter as much as user grants.

Often Confused With

Org PolicyVPC Service Controls

Common Mistakes

  • Assuming IAM only applies to human users — service accounts and workloads need explicit roles.
  • Using broad predefined roles for simplicity — increases blast radius; prefer narrow/custom roles.
  • Ignoring IAM conditions — they actively enforce constraints (time, resource, requester), not just labels.

Model Security — Protect the ML Lifecycle

Phase-specific protections (data, training, deployment, inference) to prevent poisoning, theft, and misuse of models.

Key Insight

Threats vary by lifecycle phase — use provenance & validation for training, runtime controls & monitoring for inference, and encryption + access forat

Often Confused With

Data SecurityModel Governance

Common Mistakes

  • Treating monitoring as only accuracy checks — must detect drift, adversarial inputs, and exfil attempts.
  • Relying solely on at‑rest encryption — models can be stolen or misused via APIs and inference paths.
  • Assuming cloud provider handles everything — customers must configure access, provenance, and runtime defenses.

Responsible AI Governance & Controls

Operational policies, roles, approval gates, and controls to manage privacy, bias, data quality, and explainability.

Key Insight

Governance is an operational program: role-based gates, artifact provenance, and risk‑aligned approvals — not a static checklist.

Often Confused With

ExplainabilityPrivacy/complianceMLOps

Common Mistakes

  • Treating governance as a one‑time checklist instead of ongoing roles, gates, and audits.
  • Letting legal/compliance own governance in isolation from business risk and objectives.
  • Relying on watermarking/attribution alone to eliminate misuse or legal exposure.

Explainability — Audience‑First, Approximate

Techniques and documentation (intrinsic & post‑hoc) that make model outputs understandable to specific stakeholders.

Key Insight

Explanations are approximate and role‑based — pick methods for the audience, validate fidelity, and pair with governance.

Often Confused With

InterpretabilityModel debuggingPost-hoc methods

Common Mistakes

  • Assuming explanations improve model accuracy by themselves.
  • Expecting any model to be fully and faithfully explained.
  • Treating post‑hoc explanations as the model's true reasoning without fidelity checks.

© 2026 Mocka.ai - Your Exam Preparation Partner

Google Cloud Certified Generative AI Leader Practice Questions
Access Mock Exams & Comprehensive Question Bank
Listen to Audio Podcasts
Expert summaries for Google Cloud Certified Generative AI Leader

Certification Overview

Duration:120 min
Questions:60
Passing:70%
Level:Intermediate

Cheat Sheet Content

30Key Concepts
4Exam Domains

Similar Cheat Sheets

  • PMI Certified Associate in Project Management (CAPM)® Cheat Sheet
  • PMI Professional in Business Analysis (PMI-PBA)® Cheat Sheet
  • PMI Agile Certified Practitioner (PMI-ACP)® Cheat Sheet
  • Google Cloud Professional Cloud Architect Cheat Sheet
  • Project Management Institute Portfolio Management Professional (PfMP)® Examination Cheat Sheet
  • Google Cloud Security Operations Engineer Exam Cheat Sheet
Mocka logoMocka

© 2026 Mocka. Practice for what's next.

Product

  • Browse Certifications
  • How to get started

Company

  • About Us
  • Contact

Legal

  • Terms of Service
  • Privacy Policy
  • Imprint
Follow