Trustpilot

Trustpilot processes millions of incoming reviews a day with machine learning. Most notably, it built a real-time data enrichment architecture with Google Cloud using fine-tuned Gemma models to extract structure from human-written reviews — entities, business categories, sentiment, and customer intent — in near real time.

Cost and latency. Processing millions of reviews daily under strict latency budgets makes large general-purpose models impractical. Trustpilot used larger Gemini-family teacher models with consensus annotation to generate high-quality training data, then distilled that capability into small, fast Gemma-based specialists for each task.