Google Generative AI Leader

A national telecom operator manages 2.5 million device manuals, plan FAQs, and troubleshooting guides across English, Spanish, and Korean, updated every 10 minutes from three sources: 18 TB of PDFs in Cloud Storage, public website pages, and pricing/plan tables in BigQuery; customers report poor search results on the site, and call center agents spend excessive time locating accurate, up-to-date answers—management wants a single solution that provides semantic search with grounded, citation-backed responses (RAG), respects role-based access so agents see internal documents customers cannot, integrates quickly into both the website and the agent desktop with minimal custom ML work, and delivers sub-1.5s P95 latency for common queries. Which Google Cloud solution should they use?

A national healthcare insurer stores 95,000 policy PDFs, 14,000 FAQ entries, and 2.3 million claim-processing notes across Google Cloud Storage and Confluence, and members still struggle to find answers on the portal, resulting in 180,000 monthly support chats and $2.6M in contact-center costs. They want a fully managed Google Cloud service that can index structured and unstructured content from multiple sources, provide semantic and keyword search with relevance tuning and grounding, expose APIs/embeddable UI for website and chat integration, and improve self-service CSAT while reducing support volume. Which Google Cloud product should they choose?

A city bike-sharing company logs about 50,000 trips per day and has 24 months of data with features such as trip duration, average speed, start/end stations, and start hour; the company wants to uncover natural rider segments without any labeled segment tags. Which option is an example of unsupervised machine learning in this scenario?

A city transit authority has collected 12 million time-stamped telemetry records from buses and trains, including speed (km/h), engine RPM, brake pressure (bar), and battery voltage sampled every 5 seconds over 90 days; there are no labels or categories, and the authority wants to detect potential anomalies, failures, or natural groupings of vehicle behavior using only these measurements—what type of machine learning should they use?

A national health insurer operates an LLM-powered claims helpdesk with a knowledge cutoff of December 2022 and needs to answer member questions about newly issued policy riders and regulatory bulletins that arrive at ~500 documents/day into BigQuery tables and Cloud Storage PDFs, while keeping p95 latency under 900 ms and restricting answers to sources from the last 30 days; how does retrieval-augmented generation (RAG) overcome cutoff-induced hallucinations in this setting?