Accelerate the customs clearance process with Generative AI

The Client
With a strong presence in over 90 countries, numerous local agencies, and several branches across Europe, the client is a European leader in designing and developing cutting-edge solutions in the HVAC industry.

The Challenge
The main challenge was to find a solution capable of accelerating customs clearance processes, which typically required long processing times, difficulty in maintaining consistent quality standards, and a high risk of human error, as well as limiting the ability to scale the process efficiently.

The objective was to optimize a business process previously managed manually, which involved significant time and resource consumption, aiming to deliver results in minutes that used to require several hours of work.

The key elements we focused on to develop the Generative AI-based solution were:

• High document complexity: many data points to identify, variable content, and critical entities requiring precise recognition;
• Reliability and control requirements: automation without compromising quality and traceability of information and operations;
• Cost control and optimization: need to balance efficiency with budget constraints through a serverless-first approach;
• Integration challenges: the need to insert extracted data into existing company systems in a consistent and automated way.

The Solution
Omnys developed a set of APIs on AWS leveraging the generative artificial intelligence capabilities of Amazon Bedrock. The system processes data asynchronously and efficiently using Lambda functions, making the architecture lean, scalable, and automated.

When a new document arrives, the system automatically activates the generative model, which extracts and identifies with very high precision all relevant entities (such as codes, descriptions, values, and references), even from complex or unstructured formats.

Compared to traditional OCR approaches, the solution we implemented offers superior semantic understanding and much higher adaptability.

Key advantages include:

• Overcoming the limits of traditional OCR: increased accuracy and flexibility on complex, non-standard formats with no need for manual configuration or templating, resulting in faster, more streamlined document processing;
• Cost efficiency: strategic use of AWS resources to reduce financial impact while maintaining maximum reliability;
• Scalability and flexibility: a solution that can easily adapt to evolving business needs and growing workloads;
• Easy integration with existing systems: structured output ready to feed into business workflows (REST APIs), ensuring a smooth, fast process from data acquisition to management.

OCR & Amazon Bedrock
For this client, we fully leveraged Amazon Bedrock to optimize complex OCR processes.

Unlike traditional approaches, often limited to mechanical, deterministic text extraction, we used LLM models configured with advanced prompt engineering techniques and, where needed, fine-tuning based on the client's specific data to guide the semantic interpretation of documents.
This enables the model not just to read content but to understand context, recognize relevant entities, and identify relationships between data—even with complex layouts or information spread across multiple pages.

The platform is highly adaptable and capable of generalizing the extraction process across various document types (invoices, delivery notes, technical forms, certifications), while maintaining high levels of precision and reliability.

Results and Impact
These are the tangible results achieved:

• Extremely high accuracy in entity extraction and recognition, made possible through the use of Amazon Bedrock, delivering a level of reliability previously unattainable with traditional solutions;
• Additionally, a significant reduction in processing time, moving from a semi-manual approach to a fully automated one, capable of optimizing the entire process in terms of efficiency and speed.

Thanks to this combination of technology and expertise, we are now able to take non-uniform data obtained through OCR and transform it into intelligent, automated digital processes that integrate seamlessly with existing enterprise systems.