There are several powerful generative artificial intelligence chatbots available on the net, as Gemini, ChatGPT, Code Whisperer, BingAI. These solutions provide a high-quality answer on publicly available data. When it comes to more specialized and precise requirements, these solutions are limited based on the availability of raw information.
Our 4IT AI solution has the objective to integrate to your IT landscape by using public large language models (MLL) to provide precise answers based on reliable data.
Let’s illustrate the concept.
We are a construction material seller or a construction company. Our college needs to answer following question of his client :
Which type of panels are required for “Top Colours extrrra L”?
The current MVP was built for a non-profit international organization having large panel of internal and external data sources. The MVP can be adapted to any industries assuming a proper data management system.
The application architecture assumes, that the 4IT AI solution is integrated to existing user interfaces (as a POS system) to reduce change management effort. Business users can talk to their data in a normal human language. The 4IT AI solution takes over the parsing, enrichment, transformation of these user requests into prompts or into codes. The AI solution is sourced by controlled, focused and reliable information. Public (non controlled) information can be also allowed if required.
The interpretation of the clean prompts and codes is done by an existing LLM Model.
The picture below presents the high-level technical architecture. In this scenario the AI solution is hosted within the VLAN of the company to enable access to internal data sources. A cloud-based solution is also possible, when controlled data can be accessed by secured APIs. The external LLM is not feed and trained with internal data. Only a subset of data are transferred to the external model, it van be vectorized, non vectorized or a code.
MVP illustration
The existing MVP (Minimum Viable Product) user interface has also the capacity return the results as a dynamic table or as an interactive chart.