

Who this is for
- Developers who need to stably extract business fields
- Engineer for automated data processing
What problem does this article solve?
Relying on a single generation cannot guarantee that the fields are complete, and the cost of manual repair is high.
Key concepts
- schema check
- Failure attribution
- Repair Prompt
- Maximum retry
- Review the bottom line
Walkthrough
- Parse the model output
- Capture JSONDecodeError
- Summarize missing fields
- Initiate a repair request or local mock
- Save failure samples
Python code example
The companion demo is generated in:
cd examples/llm-guides/010_json_validation_retry
python demo.py
The example follows two principles:
- From
OPENAI_API_KEY, load the key at runtime to avoid hard-coding it. - Without an API key, the example runs in mock mode so you can follow the workflow.
The essential call looks like this:
import os
api_key = os.getenv("OPENAI_API_KEY")
if not api_key:
print("未配置 OPENAI_API_KEY,进入 mock 模式。")
else:
# 真实项目中在这里初始化 SDK 客户端并发起请求。
print("已读取 OPENAI_API_KEY,可以调用模型。")
Common pitfalls
- Publishing after a single successful demo run leaves out error handling, logging and human review.
- Including API keys, real file paths, account details or business data in published articles or code.
- Readers can struggle to reproduce an example if it is unclear which parts are mocked and which run in production.
- Citing an API or framework feature without checking its documentation can mislead readers when versions change.
Summary
- LLM Guide 10 covers automatic repair when JSON output fails. A single generation can omit required fields, and fixing those omissions by hand is costly.
- Make the workflow reproducible, reversible and easy to review. One good generation tells you little about how it will hold up in use.
- Before publishing, check current documentation for model capabilities, API parameters and third-party frameworks.
CSDN tags
JSON, Python, large model, data quality, automation
WeChat article summary
The guide shows how to repair failed structured outputs automatically in data extraction and business automation workflows.
Alternative title for Zhihu
- What should I do if the JSON fails to output the large model output?
- How to design retry closed loop of structured output?
- How to reduce manual repair of LLM data extraction?
References and further reading
- Check the OpenAI documentation for your model, SDK and account permissions.
- Python documentation on environment variables, virtual environments and exception handling.
- Consult the documentation for the tools used in the example, such as LangChain, LlamaIndex, Haystack or the vector database.