Recipe

Recipe: Sentiment analysis at scale

Process millions of reviews, comments, and support tickets through a production-grade sentiment pipeline built on Meridian's inference fabric.

Batch inferenceNLPStreaming

Overview

This recipe walks through standing up a horizontally-scalable sentiment classifier. You'll ingest raw text from Kafka, classify polarity and intensity via Meridian endpoints, and write structured results to ClickHouse for real-time dashboards.

Prerequisites

Meridian API key with inference:write scope
Kafka cluster (MSK or Confluent) with topic raw-reviews
ClickHouse instance for result storage

Pipeline architecture

Kafka→Worker pool→Meridian API→ClickHouse

Workers consume batches of 100 messages, call Meridian's classify endpoint, and flush results every 5 seconds.

Next steps

Deploy the worker pool on ECS or Kubernetes with auto-scaling triggered by Kafka consumer lag. Monitor throughput and latency via the Meridian dashboard at /dashboard.