AIF-C01: App Layer Gen AI

Amazon Kendra

Intelligent Search; enterprise document indexing; multi-source unstructured query

Amazon Lex

Conversational Interface; chatbot face; text/voice dialog flow; user intents

Amazon A2I (Augmented AI)

Human-in-the-loop (HITL) workflows; human review for low-confidence ML predictions; compliance auditing

Amazon Comprehend

Natural Language Processing (NLP) API; sentiment analysis; entity extraction; PII detection & redaction; topic modeling

Amazon Transcribe

Speech-to-Text (STT); audio/live speech to text conversion; speaker diarization (who spoke when); custom vocabulary

Amazon Translate

Neural Machine Translation (NMT); language-to-language text translation; real-time/batch translation

Amazon Polly

Text-to-Speech (TTS); converts written text into lifelike speech audio; SSML support; neural voices

Amazon Rekognition

Computer Vision (CV) API; image & video analysis; facial recognition/analysis; content moderation; object/text detection

Amazon Textract

Advanced OCR & document analysis; extracts text/tables/forms from PDFs/scans; preserves document structure

Amazon Q

Turnkey Enterprise AI assistant (Q Business = secure internal search over S3/Confluence; Q Developer = code gen/IDE help)

Amazon Fraud Detector

Automated fraud prevention; fully managed service that uses ML to identify online payment fraud, fake account creation, & trial abuse

Amazon CodeGuru Security

Static Application Security Testing (SAST); uses ML to detect code vulnerabilities, hardcoded secrets, and security flaws in source code

Amazon CodeGuru Profiler

Application performance optimization; uses ML to identify expensive code lines, CPU bottlenecks, and memory leaks in runtime applications

Prompt Engineering

Zero-shot; Few-shot; Chain-of-Thought (CoT); In-context learning; No model weight changes; Fastest & cheapest customization

Fine-Tuning

Domain customization; Updates model weights; Requires labeled datasets; High compute cost; Permanent behavioral change; deeo specialized terminology

Instruction Fine-Tuning

Adapts FMs to follow specific commands/tasks; Uses prompt-response dataset pairs; Improves task alignment

Continued Pre-training

Unlabeled domain data; Adapts base FM to new specialized vocabulary/industry jargon before task tuning

PEFT (Parameter-Efficient Fine-Tuning)

Updates only a small subset of weights (e.g., LoRA); Dramatically reduces memory/GPU compute overhead

Zero-Shot Prompting

Direct instruction with NO examples provided; Relies entirely on pre-trained baseline knowledge

Few-Shot Prompting

Provides 2–5 input/output examples inside the prompt to demonstrate target format, tone, or response structure

Chain-of-Thought (CoT) Prompting

Asks model to "think step-by-step"; Deconstructs complex reasoning, logic, or math into sequential steps

Amazon Fraud Detector

Fully Managed Fraud Detection; uses machine learning and automated rules to flag online fraud like fake accounts and payment fraud

AWS Data Exchange

Data Marketplace; simplifies finding; subscribing to; and integrating third-party datasets in the cloud

AWS Elastic Inference

Cost-Optimized GPU Acceleration; attaches low-cost GPU acceleration to Amazon EC2 and SageMaker instances to reduce deep learning inference costs