Entity Recognition with LLMs: A Step‑by‑Step Guide

Описание и применение

Пошаговый рецепт автоматизации извлечения именованных сущностей из текста с помощью LLM, используя Oxlo.ai, включая конфигурацию клиента, написание системного запроса, функцию извлечения, тестирование и пакетную обработку.

Задача

Extract named entities (people, organizations, locations, dates) from unstructured text without maintaining separate NLP models, while controlling token costs.

Результат

The pipeline produces a JSON object listing all named entities per document, with support for batching to reduce per‑request costs.

Рабочий процесс

• Configure OpenAI client to point at Oxlo.ai endpoint
• Write a system prompt specifying entity types and JSON schema
• Create an extraction function that calls the LLM with JSON mode and parses the result
• Test the function on a realistic document and inspect output
• Extend the prompt for batch processing of multiple documents
• Wrap the batch function to handle lists of texts
• Run the script locally, setting the API key and executing the pipeline
• Optional: expose the extraction function via a FastAPI endpoint for external use

Доказательства

We are going to build a named entity recognition pipeline that turns unstructured text into structured JSON using an LLM.

Источник

Because Oxlo.ai charges a flat rate per request, you can feed it long documents or batch multiple inputs without watching token costs climb.

Источник

The system prompt tells the model exactly which entities to extract and what schema to return.

Источник