Finansi.bg Leveraged AI to Transform Financial Data Processing, Achieving 250k Users in Just One Year

Finansi.bg faced the formidable task of developing a platform that could efficiently extract and present critical financial data from a vast and diverse set of documents. The primary challenge was dealing with inconsistently structured data across 2 million documents, where not all pages held relevant information. This required a sophisticated solution to accurately sift through the noise and extract only the most pertinent data. Additionally, the platform needed to maintain high Google rankings, deliver exceptional site speed, and ensure a seamless user experience across all major browsers, all while controlling costs in processing and managing this extensive data.

To address these challenges, we employed a data-driven architecture combining advanced AI models, semantic retrieval and automated data-processing pipelines. The platform processes more than 2 million document records, identifying the relevant financial information hidden across inconsistently structured company reports.

The AI architecture combines Retrieval-Augmented Generation (RAG), Voyage AI embeddings and ChromaDB for semantic document retrieval with Anthropic Claude models for intelligent document analysis and structured financial-data extraction. Relevant document content is identified and supplied to Claude, which interprets and extracts the required financial information into standardized outputs. Automated validation and exception-detection mechanisms ensure that uncertain cases are routed for additional review.

This architecture enabled Finansi.bg to process financial documents at a scale and speed that would have been economically impractical using manual processing alone.

The platform combines semantic classification with Claude-powered document intelligence. Voyage AI embeddings and ChromaDB identify and retrieve relevant document pages, while Claude analyzes the selected content and extracts the required financial information into standardized structured outputs. An automated pipeline sanitizes and normalizes the data before and after model processing, while additional algorithms validate the resulting data.

Throughout the process, error detection algorithms flagged documents or pages for manual review, ensuring the highest level of data integrity. Out of all processed documents (over than 2 million), fewer than 8k required human intervention, highlighting the efficiency and accuracy of the automated system.

Our innovative approach reduced the expected data processing costs by 80% through the application of various resource optimization techniques, making the solution highly cost-efficient. By calculating the wages of manual data processors, it was evident that handling this volume manually would have been prohibitively expensive and time-consuming.

In contrast, our AI-driven solution slashed costs by 99.1%, reducing the expense from an estimated 2 million euros to just 18k euros. This automation accomplished in a fraction of the time what would have taken a team of employees over a year to process manually. This not only made the project financially viable but also ensured timely and scalable data handling for ongoing operations.

The team achieved a 98-speed rating by leveraging modern tools like Next.js, React and TailwindCSS, optimizing the platform for rapid performance across all major browsers. This focus on cross-browser compatibility ensures a seamless and responsive experience on any device or browser. The result is a highly efficient platform that exceeds industry standards for speed and usability, driving user engagement and satisfaction.