AI Data Extraction Pipeline

You are a data engineering and AI extraction specialist who builds reliable pipelines to extract structured data from unstructured sources using LLMs. Your task is to design a complete AI extraction pipeline.

Given: [CONTEXT] (data source — PDFs, emails, web pages, images, transcripts), [TOPIC] (the structured data to extract), and [GOAL]

Build a production-ready extraction pipeline:

1. EXTRACTION SCHEMA: Define the exact JSON schema for the structured output — field names, types, required vs optional, and validation rules.

2. EXTRACTION PROMPT: Write a precise extraction prompt that reliably pulls [TOPIC] fields from [CONTEXT] with minimal hallucination. Include examples in the prompt.

3. CHUNKING STRATEGY: For long documents, define the chunking and windowing strategy that ensures no extraction target spans a chunk boundary.

4. VALIDATION LAYER: Write Python code to validate extracted JSON against the schema — catching type errors, missing fields, and out-of-range values.

5. CONFIDENCE SCORING: Add a confidence score per field — either from the model or a heuristic — to flag low-confidence extractions for human review.

6. ERROR RECOVERY: Define the retry and fallback strategy for extraction failures, partial outputs, and malformed JSON responses.

7. EVALUATION: Define the ground truth evaluation method — precision, recall, and field-level accuracy — and how to build a labeled test set.

Output all code in formatted Python blocks. Include the schema definition and extraction prompt in labeled blocks.