Ntombizakhona Mabaso

for AWS Community Builders

Posted on Jan 27

Technologies And Concepts: Cheat Sheet for AI Practitioner (AIF-C01)

#ai #aws #aipractitioner #cloud

🤖 Exam Guide: AI Practitioner
Technologies And Concepts Cheat Sheet
📘Cheat Sheet

Note: Unlike the CLF-C02 exam guide, AIF-C01 does not include a dedicated "Technologies & Concepts" section. Instead, services and concepts are embedded within task statements across all five domains.

This cheat sheet consolidates that information into a compact, exam-aligned reference organized by domain—designed for quick review and efficient study.

📖 Domain 1

Fundamentals of AI & ML

1.1 Key Terms

#	Term	Definition
1	AI	Broad field of machine intelligence
2	ML	AI that learns patterns from data
3	Deep Learning	ML using multi-layer neural networks
4	Model	Trained artifact that maps inputs → outputs
5	Algorithm	Method for learning/inference (e.g., regression, trees, neural nets)
6	Training vs Inference	Learning parameters vs using the model to predict/generate
7	Fit	Underfitting vs overfitting
8	Bias & Fairness	Systematic skew and its impact on groups/outcomes

1.2 Learning Types

#	Type	Description
1	Supervised	Uses labeled data: classification, regression
2	Unsupervised	Uses unlabeled data: clustering, dimensionality reduction
3	Reinforcement Learning	Agent maximizes reward via trial/error

1.3 Inference Types

#	Type	Use Case
1	Real-time	Low latency per request
2	Batch	High-throughput scoring on a schedule

1.4 Data Types

Structured (tables) vs Unstructured (text/images/audio)
Common formats: Tabular, Time-series, Image, Text
Label status: Labeled vs Unlabeled

🧠 Domain 2

Fundamentals of Generative AI

2.1 Foundational GenAI Vocabulary

#	Concept	What It Means
1	Tokens	Units the model reads/writes; cost + limits often token-based
2	Context Window	Max tokens model can consider at once (input + output)
3	Chunking	Splitting large text into smaller pieces (often for RAG)
4	Embeddings	Numeric representations of meaning
5	Vectors	Arrays of numbers (embeddings) used for similarity search
6	Prompt Engineering	Designing instructions/context/examples to steer outputs
7	Transformers / LLMs	Architecture powering many text models (attention-based)
8	Foundation Models (FMs)	Large general-purpose models adaptable to many tasks
9	Multimodal Models	Handle multiple modalities (e.g., text + image)
10	Diffusion Models	Commonly used for image generation

2.2 Common GenAI Use Cases

Summarization
Q&A Assistants
Translation
Code Generation
Customer Service Agents
Semantic Search
Recommendations
Image/Audio/Video Generation

2.3 FM Lifecycle

Data Selection → Model Selection → Pre-training → Fine-tuning → Evaluation → Deployment → Feedback

🔧 Domain 3

Applications of Foundation Models

3.1 Model Selection Criteria

#	Factor	Why It Matters
1	Cost	Often token-based pricing
2	Latency	Response time requirements
3	Throughput	Request volume capacity
4	Model Size/Capability	Balance between power and efficiency
5	Multilingual	Language support needs
6	Modality	Text, image, audio, multimodal
7	Context Length	Input/output token limits
8	Customization Options	Fine-tuning, prompting flexibility
9	Constraints/Compliance	Regulatory requirements
10	Prompt Caching	Efficiency for repeated prefixes

3.2 Inference Parameters (Behavior Control)

Temperature: Higher = more random/creative, Lower = more consistent
Max Output Tokens: Caps response length
Other sampling settings vary by provider

3.3 RAG (Retrieval Augmented Generation)

Aspect	Details
Definition	Retrieve relevant content from a knowledge store and provide it to the FM to generate grounded answers
Why Use It	Reduces hallucinations, keeps answers current, avoids retraining for new docs
AWS Example	Amazon Bedrock Knowledge Bases

3.4 Vector Storage Options on AWS

Amazon OpenSearch Service (search + vector similarity)
Amazon Aurora
Amazon RDS for PostgreSQL
Amazon Neptune (graph-driven retrieval patterns)
Amazon DocumentDB (with MongoDB compatibility)

3.5 Customization Tradeoffs

Approach	Speed	Cost	Consistency	Best For
In-context learning (prompting/few-shot)	⚡ Fastest	💰 More tokens	⚠️ Variable	Quick iterations
RAG	⚡ Fast	💰💰 Retrieval + vectors	✅ Good	Private/updating knowledge
Fine-tuning	⏱️ Moderate	💰💰💰 Training + maintenance	✅✅ Best	Consistent behavior/style
Pre-training	⏱️⏱️ Slowest	💰💰💰💰 Extremely expensive	✅✅ Custom	Rare/large orgs only
Distillation	⚡ Fast (inference)	💰💰 Training	✅ Good	Reduce cost/latency

3.6 Agents (Multi-Step Automation)

Role: LLM plans/executes multi-step tasks and calls tools/APIs
AWS Example: Agents for Amazon Bedrock
Related Concepts: Agentic AI, Model Context Protocol (MCP)

3.7 Prompt Engineering

1 Techniques

Zero-shot: no examples
One-shot / Few-shot: with examples
Chain-of-thought: step-by-step reasoning
Prompt Templates: standardized with variables

2 Key Constructs

Instruction
Context
Negative prompts
Prompt routing

3 Best Practices

Be specific and concise
Define output format (JSON/table)
Use examples when needed
Version and test prompts
Add guardrails in prompts

4 Risks

Prompt Injection/Hijacking: especially via retrieved docs in RAG
Jailbreaking
Poisoning: bad/malicious content in sources
Exposure/Leakage system prompt or sensitive data

3.8 Evaluation Methods

1 Approaches

Human evaluation
Benchmark datasets
Amazon Bedrock Model Evaluation (managed)

2 Metrics

ROUGE: Summarization
BLEU: Translation
BERTScore: Semantic similarity

3 System-Level Evaluation

RAG: Retrieval quality + grounded answer rate + citation accuracy
Agents/Workflows: Task completion rate, tool-call correctness, cost/latency, safety

⚖️ Domain 4

Responsible AI

4.1 Responsible AI Features

Feature	What It Means
Bias	Systematic unfairness
Fairness	Equitable outcomes across groups
Inclusivity	Works for diverse users
Robustness	Reliable under variability
Safety	Prevents harmful outputs
Veracity	Truthfulness and grounding
Hallucination Awareness	Understanding model limitations

4.2 Dataset Characteristics for Responsibility

Diversity and inclusivity
Curated sources
Balanced representation
High label quality
Representativeness of real-world use

4.3 Bias/Variance Effects

Condition	Result	Impact
High Bias	Underfitting	Too simple, poor performance overall
High Variance	Overfitting	Too sensitive to training data, poor generalization
Subgroup Impact	Performance gaps	Different outcomes across demographic groups

4.4 AWS Tools for Responsible AI

Tool	Purpose
Guardrails for Amazon Bedrock	Policy/safety controls for GenAI
Amazon SageMaker Clarify	Bias detection + explainability
Amazon SageMaker Model Monitor	Monitor drift/data quality over time
Amazon Augmented AI (A2I)	Human review workflows
SageMaker Model Cards	Documentation for transparency

4.5 Transparency & Explainability

Transparent: Inherently understandable models (often simpler)
Explainable: Complex models with explanations/documentation
Tradeoff: Interpretability vs performance, transparency vs security

🔒 Domain 5

Security, Compliance, Governance

5.1 Core Security Building Blocks (AWS)

Component	Function
IAM	Roles/policies/permissions (least privilege)
Encryption	At rest and in transit
AWS PrivateLink	Private connectivity to services
Amazon Macie	Discover sensitive data in S3
Shared Responsibility Model	AWS secures cloud, you secure configs/data/apps

5.2 AI-Specific Security/Privacy Concerns

Prompt injection
Data leakage via prompts/logs
Unsafe outputs
Threat detection
Vulnerability management
Infrastructure protection

5.3 Governance Essentials

Source Citation: Document origins (trust + audit)
Data Lineage & Cataloging: Track data flow and metadata
Data Governance Strategies: Lifecycle, logging, residency, monitoring, retention

5.4 AWS Services for Audits/Governance

Service	Purpose
AWS CloudTrail	API audit logs
AWS Config	Configuration tracking/compliance rules
AWS Audit Manager	Collect evidence for audits
AWS Artifact	Download AWS compliance reports/agreements
Amazon Inspector	Vulnerability findings
AWS Trusted Advisor	Best-practice checks (security/cost)

5.5 Governance Processes

Policies and review cadence
Review strategies (human oversight, red-teaming)
Transparency standards (model cards, disclosures, citations)
Team training requirements
Framework Example: Generative AI Security Scoping Matrix

💰 Pricing & Cost Concepts

Factor	Impact
Token-based Pricing	Costs scale with input + output tokens
Latency/Responsiveness	Lower latency typically costs more
Availability/Redundancy	Multi-AZ/Region increases cost
On-demand vs Provisioned Throughput	Flexibility vs predictability tradeoff
Customization Cost Ladder	Prompting < RAG < Fine-tuning << Pre-training

📝 Common Abbreviations

Abbreviation	Full Term
LLM	Large Language Model
FM	Foundation Model
GenAI	Generative AI
RAG	Retrieval Augmented Generation
NLP	Natural Language Processing
CV	Computer Vision
PII	Personally Identifiable Information
PHI	Protected Health Information
IAM	Identity and Access Management
VPC	Virtual Private Cloud
KMS	Key Management Service
TLS	Transport Layer Security
AUC	Area Under the Curve
F1	F1 Score
ROUGE	Recall-Oriented Understudy for Gisting Evaluation
BLEU	Bilingual Evaluation Understudy

🛠️ AWS AI/GenAI Service Quick Reference

#	Service	Purpose
1	Amazon Bedrock	Use foundation models via API
2	PartyRock	Bedrock Playground for prototyping
3	Agents for Amazon Bedrock	Tool-using multi-step agents
4	Bedrock Knowledge Bases	Managed RAG building blocks
5	Amazon Q	GenAI assistant for work/dev/AWS contexts
6	Amazon SageMaker	Build/train/deploy ML, MLOps capabilities
7	SageMaker JumpStart	Pre-trained models/templates to start fast
8	SageMaker Data Wrangler	Data prep/EDA
9	SageMaker Feature Store	Consistent feature management
10	SageMaker Model Monitor	Production monitoring/drift
11	Amazon Transcribe	Speech-to-text
12	Amazon Translate	Language translation
13	Amazon Comprehend	Text analytics (entities, sentiment, etc.)
14	Amazon Lex	Chatbots/voice bots
15	Amazon Polly	Text-to-speech

⚠️ Important: Always refer to the official exam guide for the most up-to-date list of in-scope and out-of-scope services.

📚 Additional Resources

Good luck with your exam! 🚀

Top comments (3)

Black Leo • Feb 21

Successfully cleared the AWS AIP-C01A big thanks to PassCertHub.com practice questions for their accurate coverage and close alignment with the official exam objectives a valuable resource in my preparation journey.

livayey • Mar 3

The best thing I did for AIP-C01 was testing myself again and again. After completing theory, I practiced different types of questions, including some from ValidItExams.com, and that helped me manage time better in the real exam.

kogatob • Mar 3

When I first looked at the AIP-C01 syllabus, it felt a bit overwhelming. There’s a lot to cover. I focused on learning step by step and later tested myself using practice material from ValidItExams.com. That really improved my confidence before booking the exam.

Some comments may only be visible to logged-in visitors. Sign in to view all comments.