How a Global Insurance Company and Scale Are Building a Real-Time Copilot With Nemotron

Scale AI collaborated with a global insurance company to build a real-time copilot for customer service representatives using NVIDIA's Nemotron model. The AI assists during live calls by extracting case facts, recommending policy coverage, and guarding against over-promising. Post-training with supervised fine-tuning (SFT) and reinforcement learning (RL) significantly improved the model's performance, increasing fact extraction accuracy to 93.8% and coverage recommendation to 95.6%. This fine-tuning process also resulted in a 70-75% reduction in inference costs compared to the company's existing production model.