Implementing Custom NER Model for CV Parsing with Dockerization

[7matic]

• This pull request introduces an enhancement to our CV parsing. By integrating a custom trained Named Entity Recognition (NER) model, the application can now extract and process key information from CVs in PDF format.

Key Features:

• PDF Parsing: Accepts CV files in PDF format located in a specified directory.
• Dockerization: Fully dockerized, allowing for easy setup, deployment, and scaling.
• Custom NER Model: Utilizes a custom NER model to extract entities such as names, emails, phone numbers, skills, and educational backgrounds.
• Structured JSON Response: Outputs a structured JSON containing the extracted data, making it easy for further processing or integration with other systems.
• Error Handling: Error handling for file existence, readability, and format checks.
• Detailed API Documentation: Includes comprehensive documentation with examples for easy integration.

Known Issues:

• Limited Scope of NER Model: The current Named Entity Recognition (NER) model is specifically trained for English CVs from the Computer Science/IT sector.
• Model Accuracy: The accuracy of the NER model is limited due to the size of the training dataset. The model was trained on a dataset of 200 CVs.
• Skill Labeling: Skills in the CVs are labeled under a single category, rather than being separated into distinct skill types.

Fixes #6

[7matic]

Things to improve:

* Skills part of response is sometimes too long. It should include only keywords and not whole sentences.

* Example4 and 5 have mostly the same structure, yet it misses most of data from example 5.

* extract_names function may not work for people that have multiple names ex. van, von ...

Things to fix:

* Date (clock) in response headers is wrong.

* validate_file returns HTTP 400 - File does not exist, if the file is NOT a pdf. According
  to the docs. it should return 400 - File is not a PDF.

Notes:

* Maybe force the user to use a specific CV structure, like example 4, as it seems to work the best, if the model cannot be improved.

Thanks a lot for taking the time to review pull request and providing your feedback.

• Regarding the skills part of the response, you're right that it's presented as one string with keywords put together. This format is a result of the limitations imposed by the training data for the NER model. Prepared dataset labeled skills together, e.g., 'Python JavaScript Django Docker C Cpp', so results are labeled in the same manner. Unfortunately, we can't easily modify this aspect.
• You've suggested enforcing a specific CV structure like Example 4, which has shown better performance. This approach might introduce unnecessary complexities and overhead for our users; we aim to keep the user experience as smooth as possible.
• Regarding the "validate_file" function, I will update it to align with the documentation.
• The accurately extracted data, such as first names, last names, email addresses, and telephone numbers, can be utilized for display on user profiles. Additionally, data that is more likely to be incorrect (incorrect as in being labeled to wrong category by the NER model or skills being concatenated in a single string), will still be saved in the database. This data will be valuable for other tasks, such as matching resumes to jobs or filtering candidates based on job descriptions.