publications

2025

1. arXiv

   TweakLLM: A Routing Architecture for Dynamic Tailoring of Cached Responses

   Muhammad Taha Cheema and others

   arXiv preprint arXiv:2507.23674. Advised by Dr. Zafar Ayyub Qazi and Dr. Ihsan Ayyub Qazi at LUMS , 2025

   Abs arXiv Bib

   An adaptive caching architecture for LLM systems that uses a lightweight model to dynamically tailor cached responses to similar user queries. By integrating semantic caching with Milvus as a vector database, the system reduces response generation latency and cuts compute costs by up to 68% on real-world user-LLM interaction datasets.

   @article{cheema2025tweakllm,
     title = {TweakLLM: A Routing Architecture for Dynamic Tailoring of Cached Responses},
     author = {Cheema, Muhammad Taha and others},
     journal = {arXiv preprint arXiv:2507.23674},
     year = {2025},
     url = {https://arxiv.org/abs/2507.23674},
   }