Document processing and intelligent search with generative AI

At our event in collaboration with AWS named "Generative AI Solutions for Business Challenges" held on June 20, 2024, Cloud Engineer János Tassy presented a highly engaging session filled with questions and insights. Here’s a summary of the 45-minute talk, with the full presentation available at the end of this article.

The problem with paper-based documents

In the first part of his presentation, János explained the reasons why document digitalization is essential. Managing paper-based documents requires large physical storage space, consumes additional work hours, and poses numerous challenges such as:

• Difficult searchability
• Risks of fire, water damage, and physical degradation
• Significant security concerns

The issue is far from minor. According to an April 2024 report by Whale, a software development company, 45% of small and medium-sized businesses still use paper-based documentation. There is plenty of work to be done in this area. The question is: how much can AI help? As János demonstrated, the answer is a lot.

Using AI for digital transformation

When it comes to digitalizing paper-based documents and managing them thereafter, AI offers clear advantages, including:

• Time savings
• Increased efficiency
• Higher accuracy and reliability
• Scalability and flexibility
• Advanced analytics capabilities

János also touched on challenges in traditional document management and emphasized the need for metadata extraction and establishing a RAG (Retrieval-Augmented Generation) pipeline. This set the stage for the practical demonstration.

From PDF to searchable text

The session highlighted the role of Amazon Textract in optical character recognition (OCR). This machine learning-powered tool extracts text from printed or handwritten documents, even from unstructured layouts such as PDFs found online. Textract supports tasks like:

• Handwriting recognition
• Extraction of forms and tables
• Detection of layout elements and signatures
• Data extraction based on specific questions

For organizations dealing with diverse document formats, Amazon Bedrock offers a solution. Using prompting, users can create "Agents" to unify and standardize large collections of structurally different PDFs. This enables:

• Fully managed RAG workflows
• Secure connections to databases
• Retrieval of relevant information

Another Amazon product, S3, facilitates efficient document archiving. It introduces features such as access control and version tracking, which are impossible with paper-based processes.

Demonstration and cost insights

János demonstrated these concepts through a real-life example, illustrating the seamless transition from theory to practice. He also provided insights into the costs of initiating such a digital transformation process.

What’s next?

The application featured in János’s case study will soon be available to clients through their AWS accounts on the AWS Marketplace.

Watch János Tassy’s full presentation here: