The paradigm of Stream Processing has greatly matured over the last decade in its relationship with real-time AI, and in its use in tools related to domains such as healthcare and financial analysis. A central element of Stream Processing is the notion of Continuous Query, through which it’s possible to define tasks that monitor a constant and potentially unbounded stream of data [1]. To express Continuous Queries, applications can rely on two main interfaces: declarative and dataflow-based [2], which, however, remain mostly inaccessible without significant expertise in the field of query creation and optimization. In this scenario, a possible solution is represented by Natural Language Interfaces, which could help in bridging the gap between the continuous querying interface and the final user.

Natural Language Interfaces are based on the usage of dedicated large language models to translate the requests of the user, expressed through natural language, into technical languages; a notable example of this application is Text-to-SQL models, which have shown to be a promising technology for composing and optimizing SQL queries written in natural language [3].

The objective of this project is threefold: to simplify the maintenance and creation of streaming data pipelines using Text-to-Stream Natural Language Interfaces, to generate schema and pipeline evolution according to error logs and reports, and lastly, to optimize existing streaming pipelines in the fields of finance based on LLM-guided semantic optimization. 

[1] Angela Bonifati et Riccardo Tommasini. “An Overview of Continuous Querying in (Modern) Data Systems”. In : Companion of the 2024 International Conference on Management of Data, SIGMOD/PODS 2024, Santiago AA, Chile. Sous la dir. de Pablo Barcel´o, Nayat S´anchez Pi, Alexandra Meliou et S. Sudarshan. ACM, 2024. doi : 10.1145/3626246.3654679.
[2] Angela Bonifati et Riccardo Tommasini. “An overview of continuous querying in (modern) data systems”. In : Companion of the 2024 International Conference on Management of Data. 2024, p. 605-612.
[3] Kaiwen Chen, Yueting Chen, Nick Koudas et Xiaohui Yu. “Reliable text-to-sql with adaptive abstention”. In : Proceedings of the ACM on Management of Data 3.1 (2025), p. 1-30.