Asaad Cheema

Architected a Federated Meta-Learning framework, reducing training time by 90% and enabling convergence in 5–10 epochs using just 10% data. Deployed models improving anomaly detection by 15% and cutting costs by 25%. Built a topology-aware solution that reduced wireless data transmission by 70%. Developed a Human-in-the-Loop diagnostic system using LLMs, boosting prediction interpretability by 20–40%.

Researched graph network robustness under evolving structures and benchmarked parameter-efficient fine-tuning (PEFT) methods for LLMs. Demonstrated that LoRA reduces trainable parameters by over 95% while preserving full performance, enabling deployment on consumer-grade hardware.