The initial situation

With the rise of ChatGPT, many companies are wondering how artificial intelligence can create meaningful competitive advantages.

This is also the case with Validatis GmbH, a FinTech specialized in corporate data. Validatis has an extensive database, including many financial statements, annual reports and other documents in unstructured data formats. This data is essential for creating structured company profiles, answering queries from banks in the lending process and evaluating reports on ESG measures.

At present, however, extracting relevant information from this data and making it usable still requires a great deal of expert work.

The solution approach

In an initial concept sprint with Sidestream, Validatis considered how AI could add value to their business. To answer this question, the idea of an AI tool crystallized, which can be used to test how well AI can solve different problems. Examples include answering questions or structuring data based on existing company data. As another requirement, data sovereignty in particular plays a crucial role, so the outcome trade-off includes deciding between open source and commercial models. In other cases, a commercial model may deliver better results.

Validatis therefore needed a tool to apply different models to the data and compare them per situation in real operation. In this way, the appropriate model can be selected for each situation and the optimal AI assistant can be developed in the future.

Together with Sidestream's AI team, Validatis developed an evaluation tool that does just that:

• Import and preparation of internal data sources
• Application of various AI models such as GPT-3 to this data.
• The AI is asked specific questions, such as "Extract for me the sales figures of the Sidestream company from the last 3 fiscal years."
• The responses of the AI are assessed and evaluated

In this way, Validatis can test whether and how AI can help them tap into unstructured data sources and access the information they contain more efficiently.

Collaboration and technologies

The close collaboration between Sidestream and Validatis in an agile project workflow with weekly reviews made it possible to quickly understand for which use cases and business processes Validatis should profitably apply AI. By using modern technologies such as Python, Streamlit and AI libraries from HuggingFace, the tool could be implemented in a short time.