Method
LLM cascade
Try a cheap model first, escalate on low confidence.
9 cases from 8 companies apply LLM cascade. 4 report results in production.
Typical tasks: Reduce ML cost, Evaluate models, Catch fraud and abuse, Moderate content at scale and 3 more. Companies: DoorDash, Netflix, Zomato, VTB, Instacart, Walmart and 2 more. Often combined with: Few-shot prompting, LLM-as-a-judge, Threshold selection, Text classification, Hallucination detection and 32 more.
| Fact-Checking at Scale: Building a More Accurate and Efficient Synopsis Factuality System | media-streaming | 2026-11 | |
| Building Gavel: AI evaluation infrastructure across Eternal | food-delivery | 2026-10 | |
| Encoder vs LLM-as-a-Judge in Guardrails: Architecture Details and Trade-offs | general | 2026-05 | |
| An Automated Request Handling Service for SMBs | banking | 2026-05 | |
| SafeChat: DoorDash's AI-powered safety feature | food-delivery | 2025-10 | |
| Scaling Catalog Attribute Extraction with Multi-modal LLMs | e-commerce | 2025-08 | |
Attributor, or How an AI Agent Identifies the Advertiser of Digital Campaigns![]() | ![]() | adtech | 2025-05 |
| Using Predictive and Gen AI to Improve Product Categorization at Walmart | e-commerce | 2024-07 | |
| Building DoorDash's product knowledge graph with large language models | food-delivery | 2024-04 |
