University researcher Aleksi Knuutila is joining HSSH in February. He is leading the ReliaCode project, funded by the Finnish Research Council and Business Finland, studying what kind of infrastructure trustworthy AI requires for analytical workflows in organisational settings.
Large language models (LLMs) have substantially changed some areas of work, particularly software development. Their adoption in data processing and analytics has been more limited. A commonly identified constraint is reliability and auditability: analytical workflows often have low tolerance for errors, and results typically require reproducible verification. These requirements are challenging, given that LLM outputs are non-deterministic.
The ReliaCode project investigates how LLMs could support analytical work while maintaining reliability and traceability. It will evaluate approaches such as integrating LLMs with deterministic tools for specific analytical tasks, and developing methods to verify that generated implementations correspond to user intent and satisfy relevant constraints. The work will be conducted in collaboration with pilot organisations and will also examine potential routes to commercialisation.