QuestionQ36

VerifiedEvaluating and deploying agentic workflows

You are building a social media monitoring agent pipeline in Google Cloud that analyzes five million incoming customer feedback posts per day. The pipeline must classify each post into one of three sentiment categories: positive, negative, or neutral. It must also generate a one-sentence summary for each post. You need to configure the language model to minimize inference cost and latency while maintaining high classification accuracy.

What should you do?

Explanation

A fine-tuned small language model can be adapted to sentiment classification and concise summarization while using fewer resources than a larger model. High-throughput batch processing suits the volume of five million posts per day and can reduce inference costs. However, batch processing is asynchronous and does not minimize end-to-end response latency; this approach is suitable only if delayed results are acceptable. (docs.cloud.google.com)

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!