Questions about the use of large models 关于大模型使用的问题

[UNAOJH]

Hello,

I’d like to share a few suggestions to improve the AI receipt recognition feature in EZ. These ideas are intended to enhance usability and learning, and I hope they can be considered for future updates.

1. Understanding OCR functionality

I noticed that selecting a visual understanding model is not necessary; the system can correctly recognize receipts using a text-based model like qwen3.5-flash. I’m curious whether the OCR functionality is handled by EZ itself. Understanding this would help me learn how the system integrates OCR with the LLM.

2. Logging the actual model used

It would be helpful if the logs could indicate which model is used for each recognition request. I understand this might seem redundant since the model is configured in the settings. However, as a student, seeing the actual model called in the logs would help me understand how configuration maps to execution, and it would be useful for debugging or learning purposes.

3. Batch receipt recognition

It would be very useful if AI image recognition could support batch addition of receipts. Currently, I need to enter each payment detail page, capture a screenshot, and upload it individually. Allowing batch uploads directly from a common bill overview page would save time and improve workflow efficiency.
)
For example, after batch uploading, I could scroll through multiple receipts. This would require minor adjustments to the image upload interface to support multiple images instead of just one.

Although this feature may involve some complexity, I’m confident it would be a valuable addition to EZ, making receipt management faster and more convenient, and it would also help me learn more about how the system works.

Thank you for considering these suggestions! I really appreciate all the work you’ve done on EZ, and I hope these ideas can contribute to further improvements.

第一，我注意到使用的AI模型可以不需要选择视觉理解模型直接选择文本模型依然能正常工作，例如我使用aliyun的qwen3.5-flash模型，他依然可以正常识别，那么OCR是“ez”的功能吗

第二，进行大模型识别的时候，能否在日志加一个是调用的谁的模型，我知道这个非常鸡肋，因为已经在配置文件配置了 ，但是我是学生我想学习一下如何进行配置和具体调用的是谁的模型，有时候我反其道而行之却依然调用成功，我觉得加一个日志更清晰。

第三，AI识图的时候可以批量添加的话就更好了，比如我不需要进入支付详情页然后一个一个截屏接着拖入到ez去识别，而是在一个公共的账单界面直接截屏即可批量添加
)
例如我这张图片，在批量添加之后向下滚动即可批量查看并且添加。那么在添加图片的界面也需要稍作修改，不能只支持添加单个图片

我知道我的第三个功能也许是一个挑战，但是我相信作者能够战胜，也为ez增添更多的功能

[UNAOJH]

I forgot that you can adjust the log output level and startup mode in the configuration file, I used debug and development to start and saw more detailed logs, which allowed me to solve the first and second and puzzles, and now there is a third problem for developers to consider

[mayswind]

1. ezBookkeeping does not natively support OCR. The AI image recognition feature relies on LLM with image understanding capability. The capabilities of the Qwen model you are using can be found at https://modelstudio.console.alibabacloud.com/ap-southeast-1/#/model-market/detail/qwen3.5-flash
2. Because AI generated results may contain mistakes, ezBookkeeping requires manual confirmation after each image is recognized before the record is added. This workflow is not suitable for batch processing. If you need batch operations, you can implement them using Agent Skill or MCP provided by ezBookkeeping.