This project aims to enable the real-time execution of Large Language Models (LLMs) on mobile devices with limited computational resources and memory. Prospective candidates involved in this process will systematically research and comparatively evaluate model compression techniques such as quantization, pruning, and knowledge distillation.

The resulting optimized models will be deployed to Android and iOS-specific runtime environments to establish end-to-end, on-device inference pipelines, where the trade-off between accuracy and latency will be experimentally measured. The ultimate objective of developing these models is to reliably deliver natural language processing capabilities within mobile applications integrated with Baykar’s unmanned aerial vehicles (UAVs) and field operation systems, ensuring offline functionality and low power consumption.

Criteria and Expectations

• In line with the targeted outcomes of the project, candidates are expected to possess the following competencies and interests:
• A strong interest in Natural Language Processing (NLP) and Large Language Models (LLMs), coupled with a high motivation for continuous learning in this domain.
• An eagerness to conduct research on AI model compression and optimization techniques, including quantization, pruning, and knowledge distillation.
• Familiarity with mobile application ecosystems and runtime environments for Android and/or iOS platforms.
• The ability to optimize for accuracy and latency on embedded and mobile devices constrained by limited computational resources.
• An interest in highly energy-efficient software architectures capable of functioning in operational field environments without internet connectivity.

Relevant Academic Departments

• Computer Engineering
• Artificial Intelligence Engineering / Artificial Intelligence and Data Engineering
• Software Engineering
• Information Systems Engineering
• Electrical and Electronics Engineering