There is a quite popular framework for building AI applications in Python called DsPy (https://dspy.ai/) which is open-source (https://github.com/stanfordnlp/dspy).
One of the most interesting features of this framework is MIPRO (Multiprompt Instruction PRoposal Optimizer) prompt optimizer (https://dspy.ai/api/optimizers/MIPROv2/).
On our end we recently and very successfully launched a framework for building AI agents and AI applications in Kotlin and JVM called Koog (https://docs.koog.ai/) which is also open-source (https://github.com/JetBrains/koog). Please feel free to check out some videos about it to get yourself familiar with the framework:
- Kotlin Conf 2025 presentation of Koog: https://www.youtube.com/watch?v=O8WQCrdza8E&t=1377s — the moment when it was launched
- "Buildiung AI agents with Koog" lifestream series: https://www.youtube.com/watch?v=vysVNg4IuUo&t=26s , https://www.youtube.com/watch?v=vDtnqQmiyck&t=24s
- "How to Build Scalable AI Agents with Kotlin, Ktor & Koog": https://www.youtube.com/watch?v=AGHONAx8gjQ&t=268s
- "Kotlin summit 2025 - Building AI agents in Kotlin" https://www.youtube.com/watch?v=K4O7Qz5-N_Q&t=682s
As part of this internship, your task will be to study how the MIPRO optimizer works, implement a similar system in Kotlin, and design its integration into the Koog framework. This project will give you:
Hands-on experience with ML optimization methods
Experience of API design in Kotlin
and all the real-world challenges of building production-ready tools for modern AI frameworks on the JVM and Kotlin Multiplatform
Good understanding and experience with Machine Learning
Knowledge of Java/Kotlin/other JVM language — must have!
Ability to read Python
Being able to work independently
Willingness to collaborate with the team, being proactive, and desire to potentially join our collaborative efforts to build the best open-source AI framework