Improving Agent Feedback with Multi-LoRA at Convirza

Convirza, an AI-driven software platform, has significantly enhanced its ability to evaluate customer service performance by partnering with Predibase. This collaboration allowed Convirza to tackle the challenges of processing millions of call recordings and extracting actionable insights using conversation analytics. By leveraging Predibase’s multi-LoRA serving infrastructure, Convirza achieved a 10x reduction in operational costs, an 8% improvement in F1 scores, and an 80% increase in throughput compared to OpenAI, enabling them to efficiently manage over 60 performance indicators. Convirza's previous limitations with Longformer models were addressed by adopting Predibase’s scalable infrastructure, which facilitated faster training and deployment of small language models (SLMs) and improved responsiveness to customer needs. The transition to Predibase’s platform allowed Convirza to maintain high-performance standards during variable workloads, achieving sub-second mean inference times and reducing the burden of infrastructure management, ultimately enhancing customer service quality through better agent performance evaluation.