We trained EasyDocs' invoice extraction model

Their goal was to improve the accuracy of extracting key invoice details such as VAT numbers, gross amounts, and addresses using a combination of AI models, and OCR technology.

The existing system faced several challenges, including incomplete field detection and the limited ability to learn from previous user annotations, resulting in manual adjustments that slowed down the workflow.

EasyDocs aimed to automate this process further, enabling the system to learn from user input and correctly identify fields when patterns repeat across different invoices.

Improved Accuracy in Data Extraction: We upgraded the system ability to accurately detect and extract key fields from invoices. This reduced the number of errors and manual corrections needed by users.

AI Model with Continuous Learning: The new system uses a fine-tuned Natural Language Processing (NLP) model, to identify and annotate fields such as VAT numbers and gross amounts. This model learns from user corrections, improving its accuracy over time. We provided them with all necessary configurations and support to implement the system locally without relying on external services.

Phase 1 - Week 1: Requirement Analysis: We conducted in-depth discussions with EasyDocs to understand their needs and challenges in invoice processing.

Phase 2 - Prototype Development: A prototype and demo was created to validate the approach on a sample dataset and user feedback.

Phase 3 - Model Tuning and Finetuning: We refined the OCR and NLP models based on feedback from EasyDocs, incorporating additional datasets and improving model accuracy for multilingual invoices.

Phase 4 - Testing and Validation: The solution was rigorously tested on a wide range of accurate invoices to ensure it met the target accuracy of 80% for automatic field identification.

Phase 5 - Deployment and Training: The final model was deployed on EasyDocs' servers, and their team was prepared on how to manage and update the system as needed.

Improved Field Detection Capabilities: The enhanced OCR and AI model improved field detection accuracy and expanded the range of detected fields.

Reduced Manual Intervention: With the system's learning capability, the need for manual intervention in marking fields dropped, saving substantial time and resources.

Scalable In-house Solution: The system was successfully deployed on EasyDocs' servers, providing a scalable and secure solution that could be adapted for future needs.