Architecting Low-Code AI Solutions - Quiz

Question 1 of 10

A retail company wants to implement a solution to detect inappropriate product reviews on their e-commerce platform in real-time. They need to identify reviews containing profanity, personal attacks, and spam across 15 different languages. Which Google Cloud service would be the most appropriate and cost-effective solution?

Train a custom text classification model using AutoML Natural Language Use the Natural Language API's content classification feature with custom categories Use the Natural Language API's moderateText method from the Perspective API integration Deploy a pre-trained BERT model on Vertex AI for content moderation

Explanation

The Natural Language API's moderateText method (Perspective API integration) is specifically designed for content moderation tasks including detecting toxicity, profanity, and inappropriate content across multiple languages. It's a pre-built, low-code solution that doesn't require training. Option A (AutoML) would be unnecessarily complex and expensive for this well-defined use case. Option B is incorrect because content classification is for categorizing content into predefined topics, not moderation. Option D requires more infrastructure management and ML expertise than necessary for this standard use case.

Question 2 of 10

Your healthcare organization needs to extract medication names, dosages, and medical conditions from thousands of unstructured clinical notes. The solution must comply with HIPAA requirements and maintain data residency within your VPC. What is the most appropriate approach?

Use the public Healthcare Natural Language API endpoint with API keys for authentication Deploy the Healthcare Natural Language API with VPC Service Controls and Private Service Connect Use the standard Natural Language API with entity extraction and train it on medical terminology Build a custom entity extraction model using AutoML Natural Language on Vertex AI

Explanation

The Healthcare Natural Language API with VPC Service Controls and Private Service Connect ensures data stays within your VPC perimeter and provides HIPAA-compliant medical entity extraction capabilities specifically designed for healthcare data. Option A violates data residency requirements by using public endpoints. Option C is incorrect because the standard Natural Language API lacks specialized medical knowledge and doesn't provide HIPAA compliance guarantees. Option D is unnecessarily complex when a purpose-built healthcare API exists, and would require significant medical training data and expertise.

Question 3 of 10

A media company wants to automatically generate closed captions for their video content library containing interviews with speakers who have strong regional accents. Initial tests with the Speech-to-Text API show poor accuracy. What is the best approach to improve transcription quality using low-code solutions?

Switch to using Chirp, Google's universal speech model, which better handles diverse accents and audio conditions Use the Speech-to-Text API with phrase hints containing expected terminology Create a custom speech model using Speech-to-Text API's model adaptation with audio samples Pre-process audio files with Cloud Storage audio normalization before transcription

Explanation

Chirp is Google's latest universal speech model specifically designed to handle diverse accents, dialects, and challenging audio conditions without requiring custom training. It represents the most low-code, effective solution for accent challenges. Option B (phrase hints) helps with specific terminology but doesn't address accent recognition issues. Option C requires collecting training data and model training, making it higher complexity. Option D doesn't address the core accent recognition problem and audio normalization is not a standard Cloud Storage feature for this purpose.

Question 4 of 10

Your e-commerce company needs to implement visual search functionality where customers can upload photos of products and find similar items in your catalog of 500,000 products. The solution must return results in under 200ms. Which architecture would be most appropriate?

Use Vision API to extract labels and perform text-based search in BigQuery Create vector embeddings using Vertex AI Vision Embeddings API and store them in Vertex AI Vector Search Use AutoML Vision to train a custom classification model with all product categories Use Vision API's web detection feature to find similar images on the internet

Explanation

The Vertex AI Vision Embeddings API combined with Vertex AI Vector Search is purpose-built for visual similarity search at scale with low-latency requirements. It generates dense vector representations of images and enables fast nearest-neighbor search across large catalogs. Option A loses visual similarity information by converting to text labels. Option C is impractical for 500,000 products and doesn't provide similarity search, only classification. Option D searches the internet rather than your product catalog and doesn't meet the use case requirements.

Question 5 of 10

A financial services company wants to build a chatbot that answers customer questions about their account balances, transaction history, and investment portfolios using natural language. The solution must integrate with their existing customer database and provide accurate, grounded responses. What is the most appropriate architecture?

Use Dialogflow CX with webhook fulfillment to query customer databases and return structured data Use PaLM API for Chat with function calling to invoke Cloud Functions that query customer data Fine-tune a PaLM model on customer service transcripts and deploy it on Vertex AI Use Vertex AI Search for enterprise to index customer data and provide conversational search

Explanation

PaLM API for Chat with function calling enables the model to understand user intent and invoke specific functions (via Cloud Functions) to retrieve real-time, accurate customer data from databases, then formulate natural language responses. This approach grounds responses in actual data. Option A (Dialogflow CX) works for structured, flow-based conversations but is less flexible for complex natural language understanding. Option C (fine-tuning) is expensive, complex, and the model wouldn't have access to real-time data. Option D is designed for document search rather than structured database queries and conversational interactions.

Question 6 of 10

Your company receives customer feedback emails in 40+ languages and needs to route them to appropriate departments (billing, technical support, sales) based on content. The system should handle 10,000 emails daily with minimal maintenance. What is the optimal solution?

Use Translation API to translate all emails to English, then use AutoML Natural Language for classification Use the Natural Language API's classifyText method which supports multilingual classification Create separate AutoML Natural Language models for each of the major languages Use PaLM API with a prompt that includes few-shot examples for classification

Explanation

The Natural Language API's classifyText method provides built-in multilingual support and can classify content into predefined categories without requiring translation or separate models per language. It's a low-maintenance, scalable solution. Option A adds unnecessary translation costs and latency. Option C is operationally complex, expensive, and difficult to maintain across 40+ languages. Option D (PaLM with prompts) could work but would be more expensive at scale, requires careful prompt engineering, and classifyText is purpose-built for this exact use case.

Question 7 of 10

A real estate company wants to automatically blur faces and license plates in property photos before publishing them online. They process approximately 50,000 images per day. Which solution provides the best balance of accuracy, simplicity, and cost?

Use Vision API's face detection and object detection, then apply blurring with Cloud Functions Use Vision API's Safe Search detection to identify images requiring blurring Train a custom AutoML Vision object detection model to identify faces and license plates Use Vertex AI's pre-trained object detection model with custom post-processing

Explanation

Vision API provides highly accurate face detection and can detect vehicles (including license plates) through object detection. Combined with simple image blurring logic in Cloud Functions, this is a straightforward, low-code solution. Option B (Safe Search) detects inappropriate content, not faces/license plates. Option C is unnecessarily complex and expensive when pre-built APIs already handle these common detection tasks well. Option D adds unnecessary complexity by using Vertex AI when Vision API is specifically designed for this use case.

Question 8 of 10

Your logistics company needs to process delivery proof documents that include handwritten signatures, printed text, and checkboxes in various layouts. You need to extract customer names, delivery dates, and checkbox states. What is the most effective low-code approach?

Use Vision API's text detection (OCR) and build custom parsing logic Use Document AI's specialized processor for logistics documents Train a custom AutoML Vision model for document understanding Use Document AI's Form Parser processor to extract key-value pairs and entities

Explanation

Document AI's Form Parser is specifically designed to extract structured data (key-value pairs, checkboxes, entities) from documents with varying layouts without requiring custom training. It handles handwritten and printed text effectively. Option A (Vision API OCR) only extracts text without understanding document structure or relationships. Option B assumes a pre-built logistics processor exists, which may not be the case. Option C requires significant training data and expertise when a purpose-built document understanding solution exists.

Question 9 of 10

A content creation platform wants to implement a feature where users can generate custom images from text descriptions. The solution must support safety filters to prevent inappropriate content generation and provide reproducible results. Which approach is most appropriate?

Use Vertex AI's Imagen API with safety filters enabled and specify seed values for reproducibility Use the Vision API's image annotation to validate generated images from an open-source model Deploy Stable Diffusion on Vertex AI Prediction with custom safety classifiers Use PaLM API to generate image URLs from existing stock photo services

Explanation

Vertex AI's Imagen API is Google's purpose-built text-to-image generation service that includes built-in safety filters, supports seed values for reproducibility, and is a fully managed low-code solution. Option B doesn't address image generation itself. Option C requires significantly more operational complexity, model management, and custom safety implementation. Option D doesn't actually generate images; it only retrieves existing images and limits creative possibilities.

Question 10 of 10

Your customer service team needs to analyze recorded support calls to identify: 1) customer sentiment, 2) discussed topics, 3) whether specific compliance phrases were mentioned. You have 2,000 hours of audio in 5 languages. What is the most efficient architecture using Google Cloud's low-code AI services?

Use Speech-to-Text API for transcription, then separate calls to Natural Language API for sentiment analysis, entity extraction, and custom entity detection Use Speech-to-Text API with sentiment analysis enabled, then use Natural Language API for entity and syntax analysis Use Contact Center AI Insights to get transcription, sentiment, topics, and custom highlights in a single integrated solution Use Chirp for transcription, then use PaLM API with custom prompts to extract all required information

Explanation

Contact Center AI Insights is specifically designed for analyzing customer conversations and provides integrated transcription, sentiment analysis, topic detection, and custom highlighting (for compliance phrases) in a single, purpose-built solution. This reduces integration complexity and costs. Option A requires multiple API calls and custom integration. Option B doesn't fully address topic extraction and compliance phrase detection. Option D would be more expensive and require careful prompt engineering when a specialized solution exists for this exact use case.

Welcome Back

Terms of Service & Privacy Notice

Create Account

Verify Your Phone Number

Reset Your Password

Login Required

Get in Touch

Building AI solutions by using ML APIs or foundational models