The solution – a generative AI tool that searches and summarizes legal texts
We first did a pre-study to check that there was a business case for creating an AI tool for legal professionals. We also studied what is the right balance between technological capabilities and costs.
We then created a proof of concept for Edilex Lakitieto that showed what’s possible to achieve. The user can type in a text prompt and the system will return answers from their extensive legal knowledge base, along with sources that can be checked – which is vital as the accuracy of information is of the utmost importance in the legal field.
After the proof of concept showed promise, we wanted to find the best way to create the full solution. We investigated the capabilities of different language models like OpenAI’s GPT, Google’s Gemini, and Anthropic’s Claude models. We also compared embedding models, which are used to represent real-world objects like words in ways machines can understand. The chosen models will form the basis of a future AI service for Edilex Lakitieto customers. One critical element was to find a way to validate presented information and show sources to the user – Edilex Lakitieto is the most trusted juridical information provider in Finland and any information needs to be accurate.
Because the AI field is developing rapidly and changing all the time, it's important to work in an agile way. The goal has been to build the architecture in a way that allows us to be flexible and continue to use the best solution as generative AI tools develop. We also want to build a sustainable AI tool that is relevant and trustworthy. The project has an ambitious timeline in order to meet Edilex Lakitieto’s business needs.
