AWS AI Services

AWS AI Services (Pre-trained APIs)

AI Services provide pre-trained ML models via simple APIs. No ML expertise, no training data, no model deployment — just call an API and get a result. Pay per use.

Quick Comparison

Service	What it Does	Input	Output
Rekognition	Image/video analysis	Image bytes or S3	Labels, faces, text, celebrities
Comprehend	NLP text analysis	Text or S3	Sentiment, entities, PII, topics
Polly	Text-to-speech	Text	MP3/PCM audio
Translate	Neural machine translation	Text or S3	Translated text
Transcribe	Speech-to-text	Audio or S3	Transcripts
Textract	Document extraction	Image/PDF or S3	Text, tables, forms
Lex	Chatbots	Text or voice	Intent/slot parsing
Kendra	Enterprise search	Questions	Ranked answers
Contact Lens	Contact center analytics	Audio/text	Sentiment, categories

Rekognition (Vision)

Image Analysis

import boto3
 
rekognition = boto3.client('rekognition')
 
# Detect objects and scenes
response = rekognition.detect_labels(
    Image={'S3Object': {'Bucket': 'my-bucket', 'Name': 'photo.jpg'}},
    MaxLabels=10
)
 
for label in response['Labels']:
    print(f"{label['Name']} ({label['Confidence']:.1f}%)")

Face Comparison

# Compare faces
response = rekognition.compare_faces(
    SourceImage={'S3Object': {'Bucket': 'my-bucket', 'Name': 'id-card.jpg'}},
    TargetImage={'S3Object': {'Bucket': 'my-bucket', 'Name': 'selfie.jpg'}},
    SimilarityThreshold=80
)
 
for match in response['FaceMatches']:
    print(f"Match: {match['Similarity']:.1f}%")

Video Analysis

# Start video analysis
response = rekognition.start_label_detection(
    Video={'S3Object': {'Bucket': 'my-bucket', 'Name': 'video.mp4'}},
    MinConfidence=80,
    NotificationChannel={'SNSTopicArn': 'arn:aws:sns:us-east-1:...', 'RoleArn': 'arn:aws:iam::...'}
)

Comprehend (NLP)

Sentiment Analysis

comprehend = boto3.client('comprehend')
 
# Detect sentiment
response = comprehend.detect_sentiment(
    Text="I absolutely love this product! Best purchase ever.",
    LanguageCode='en'
)
print(response['Sentiment'])  # POSITIVE
print(response['SentimentScore'])  # {POSITIVE: 0.99, ...}

Entity Detection

# Detect entities (people, places, organizations)
response = comprehend.detect_entities(
    Text="Amazon opened its new office in Seattle last week. CEO Andy Jassi announced.",
    LanguageCode='en'
)
 
for entity in response['Entities']:
    print(f"{entity['Type']}: {entity['Text']} (Score: {entity['Score']:.2f})")

PII Detection

# Detect personally identifiable information
response = comprehend.detect_pii_entities(
    Text="Customer John Smith, SSN 123-45-6789, email john@example.com",
    LanguageCode='en'
)
 
for pii in response['Entities']:
    print(f"{pii['Type']}: {pii['Score']:.2f}")

Medical Comprehend (HIPAA Eligible)

medical = boto3.client('comprehendmedical')
 
# Detect medical entities
response = medical.detect_entities_v2(
    Text="Patient has Type 2 diabetes. Takes 500mg Metformin twice daily."
)
 
for entity in response['Entities']:
    print(f"{entity['Category']}: {entity['Text']} ({entity['Type']})")

Polly (Text-to-Speech)

polly = boto3.client('polly')
 
# Synthesize speech
response = polly.synthesize_speech(
    Text="Hello! Welcome to AWS machine learning services.",
    OutputFormat='mp3',
    VoiceId='Joanna'  # Neural voice
)
 
# Save to file
with open('welcome.mp3', 'wb') as f:
    f.write(response['AudioStream'].read())

Speech Synthesis Marks (SSML)

response = polly.synthesize_speech(
    Text='<speak><prosody rate="slow">This is slow.</prosody> <break strength="strong"/> <prosody rate="fast">This is fast.</prosody></speak>',
    OutputFormat='mp3',
    VoiceId='Joanna',
    TextType='ssml'
)

Translate

translate = boto3.client('translate')
 
# Translate text
response = translate.translate_text(
    Text="Hello, how are you?",
    SourceLanguageCode='en',
    TargetLanguageCode='es'
)
print(response['TranslatedText'])  # "Hola, ¿cómo estás?"

Batch Translation

# Batch translate documents in S3
response = translate.start_text_translation_job(
    InputDataConfig={'S3Uri': 's3://my-input/batch/', 'ContentType': 'text/plain'},
    OutputDataConfig={'S3Uri': 's3://my-output/batch/'},
    SourceLanguageCode='en',
    TargetLanguageCode='es',
    DataAccessRoleArn='arn:aws:iam::123456789012:role/translate-role'
)

Transcribe (Speech-to-Text)

transcribe = boto3.client('transcribe')
 
# Start transcription job
transcribe.start_transcription_job(
    TranscriptionJobName='my-podcast',
    Media={'MediaFileUri': 's3://my-bucket/podcast.mp4'},
    MediaFormat='mp4',
    LanguageCode='en-US',
    Settings={
        'ShowSpeakerLabels': True,
        'MaxSpeakerLabels': 10,
        'VocabularyName': 'custom-vocabulary'  # Optional custom words
    }
)
 
# Get result
response = transcribe.get_transcription_job(TranscriptionJobName='my-podcast')
print(response['TranscriptionJob']['Transcript']['TranscriptFileUri'])

Real-time Transcription

# Start stream transcription (WebSocket)
transcribe.start_stream_transcription(
    LanguageCode='en-US',
    MediaSampleRateHertz=16000,
    MediaEncoding='pcm',
    VocabularyName='custom-vocabulary'
)

Textract (Document Extraction)

Text and Tables

textract = boto3.client('textract')
 
# Analyze document
response = textract.analyze_document(
    Document={'S3Object': {'Bucket': 'my-bucket', 'Name': 'invoice.pdf'}},
    FeatureTypes=['TABLES', 'FORMS']
)
 
# Extract tables
for block in response['Blocks']:
    if block['BlockType'] == 'TABLE':
        table = textract.get_table_document(TableBlockId=block['Id'])
        # Process table data

Expense Receipts

# Analyze expense
response = textract.analyze_expense(
    Document={'S3Object': {'Bucket': 'my-bucket', 'Name': 'receipt.jpg'}}
)
 
for expense_field in response['ExpenseDocuments'][0]['SummaryFields']:
    print(f"{expense_field['Type']}: {expense_field['ValueDetection']['Text']}")

Lex (Chatbots)

lex = boto3.client('lexv2-runtime')
 
# Send message
response = lex.recognize_text(
    botId='my-bot-id',
    botAliasId='my-alias',
    localeId='en-US',
    sessionId='user-123',
    text='I want to book a flight'
)
 
for message in response['messages']:
    print(f"{message['content']}")

Kendra (Enterprise Search)

kendra = boto3.client('kendra')
 
# Query
response = kendra.query(
    IndexId='my-index-id',
    QueryText='What is our return policy?'
)
 
for result in response['ResultItems']:
    print(f"{result['ScoreAttributes']['ScoreConfidence']}: {result['DocumentTitle']['Text']}")
    print(f"  {result['DocumentExcerpt']['Text'][:200]}...")

Pricing

Service	Cost
Rekognition (image)	$0.0012/image (first 1M), cheaper after
Rekognition (video)	$0.10/minute
Comprehend (sentiment)	$0.0001/character
Polly (Neural)	$0.016/1K characters
Translate	$0.000015/character
Transcribe	$0.024/15seconds(standard),$0.042/15 seconds (medical)
Textract (sync)	$0.0015/page(text),$0.015/page (forms/tables)
Lex	$0.004/utterance
Kendra	$0.25/1K queries (enterprise edition)

References

• Homepage: https://aws.amazon.com/ai/services/
• Documentation: https://docs.aws.amazon.com/rekognition/, https://docs.aws.amazon.com/comprehend/, etc.
• Pricing: https://aws.amazon.com/ai-services/pricing/

Nuggets & Gotchas

• AI Services return confidence scores — 99% confidence doesn’t mean 100% correct: Always treat AI Service outputs as probabilistic. For high-stakes decisions, add human review or validation. A 99% confidence face match is still wrong 1% of the time.
• Rekognition’s face comparison is NOT identity verification — it’s similarity scoring: Rekognition tells you two faces are 95% similar. It doesn’t tell you WHO the person is. For identity verification, use Amazon Verify or a different approach.
• Comprehend Medical is a separate service with different pricing — Comprehend (general) is NOT HIPAA eligible: If you need HIPAA-compliant NLP, use comprehendmedical endpoint, not comprehend. They’re separate APIs with different compliance certifications.
• Transcribe supports custom vocabularies but not custom models — if your domain vocabulary is niche, build a custom vocabulary: Custom vocabulary improves accuracy for domain-specific terms (medical, legal, technical). Without it, “Glucoma” gets transcribed as “Glaucoma” incorrectly.
• AI Services are eventually consistent — for the same input, you might get slightly different outputs over time: If you need deterministic outputs (for testing or compliance), be aware that AI Service outputs can vary slightly between calls for the same input.