Method
Zero-shot classification
Classifying without task-specific training via NLI or LLM prompts.
8 cases from 7 companies apply Zero-shot classification. 2 report results in production.
Typical tasks: Catch fraud and abuse, Moderate content at scale, Build an assistant over internal docs, Evaluate models and 2 more. Companies: Yelp, Raiffeisenbank, Avito, Rabota.ru, Positive Technologies and 2 more. Often combined with: Embedding similarity matching, Supervised fine-tuning, Embedding fine-tuning, Multimodal embeddings, Threshold selection and 17 more.
| Encoder vs LLM-as-a-Judge in Guardrails: Architecture Details and Trade-offs | general | 2026-05 | |
| Improving a Modular Chatbot with LLMs: Practical Experience | banking | 2025-05 | |
RAG as a Flexible Method for Creating Language Assistants![]() | ![]() | general | 2024-05 |
| Launching Automated Video Moderation | classifieds | 2024-05 | |
| Yelp’s AI pipeline for inappropriate language detection in reviews | marketplace | 2024-03 | |
| Opening the Door to the World of Machine Learning's Dark Forces: Zero-shot Classification for Ruthless Non-Personalized Recommendations | classifieds | 2023-10 | |
| Yelp Content As Embeddings | marketplace | 2023-04 | |
Phishing detection + domain experts![]() | ![]() | general | 2022-10 |

