AIF-C01 Cheat Sheet: AI Practitioner Quick Reference

AIF-C01 is vocabulary-driven: distinguish AI/ML/GenAI, know which AWS AI service does what, and know responsible-AI controls. Scan this the week before the exam.

Vocabulary you must separate

• AI = the field; ML = subset learning from data; DL = neural-net subset; GenAI = creates new content from foundation models
• Foundation model (FM) = large pre-trained model adaptable to many tasks; LLM = text-focused FM
• Inference = running a trained model; training = fitting parameters on data; fine-tuning = continued training on your data
• Embeddings = numeric vectors capturing meaning → semantic search, RAG retrieval

AWS AI services in one line

• SageMaker — build/train/deploy custom ML models; SageMaker JumpStart = pre-trained models
• Bedrock — managed foundation models via API (Claude, Llama, Titan, etc.) + Knowledge Bases + Agents + Guardrails
• Comprehend — NLP: sentiment, entities, PII; Textract — document text/tables/forms extraction
• Rekognition — image/video analysis; Polly — text→speech; Transcribe — speech→text; Translate — language translation
• Lex — chatbots; Personalize — recommendations; Forecast — time-series; Fraud Detector / Kendra (search)
• Q Business / Q Developer — enterprise assistant / coding assistant

Prompt engineering levers

• Temperature — higher = more random/creative, lower = more deterministic
• Top-p (nucleus) / top-k — sample from most probable tokens
• Zero-shot vs few-shot (examples in prompt); chain-of-thought improves reasoning
• System/context prompts set behavior; prompt injection = user input overriding instructions

RAG vs fine-tuning

• RAG — retrieve relevant docs at inference → current/factual answers without retraining; cheaper, traceable
• Fine-tuning — adjust model weights on labeled data → style/format/domain behavior; costlier, static knowledge
• Bedrock Knowledge Bases = managed RAG; Agents = tool-use orchestration

Responsible AI

• Dimensions: fairness, explainability, privacy, safety, transparency, veracity/robustness, governance
• Guardrails for Amazon Bedrock — content filters, denied topics, PII redaction, grounding checks
• Hallucination = confident false output — mitigate with RAG grounding + guardrails
• SageMaker Clarify / Model Monitor — bias detection and drift monitoring

AI security & governance

• Same AWS security stack applies: IAM for access, KMS for encryption, CloudTrail for audit, PrivateLink for private endpoints
• Data residency & retention: understand where Bedrock processes prompts; training data doesn't leave your control plane boundary per service terms
• Shared responsibility extends to AI: AWS secures the platform, you secure prompts/data/outputs