LLTutor – AI Educational Platform|

LLTutor, an AI-powered educational platform designed to deliver personalized learning using advanced language models. The goal was to build an intelligent tutor that could adapt explanations, generate exercises, and support students based on their learning needs.

To achieve this, I worked with LLaMA2, a state-of-the-art large language model. I fine-tuned the model using Python, NumPy, PyTorch, and Google Colab, taking full advantage of GPU acceleration through CUDA. Instead of doing full fine-tuning—which is computationally expensive—I implemented LoRA (Low-Rank Adaptation) and PEFT (Parameter-Efficient Fine Tuning). These techniques allowed me to fine-tune the model more efficiently by training only small adapter layers while keeping the main model weights frozen. This dramatically reduced both training time and hardware requirements.

After training, I prepared the model for lightweight deployment by converting it into GGUF format using llama.cpp. This step was important because it allowed the model to run smoothly on low-resource devices such as laptops or edge hardware without needing a full GPU server. It made the platform more accessible and practical for real-world use.

Finally, I uploaded the optimized models to the Hugging Face Hub, enabling public access, reproducibility, and collaboration. This also made it easier for others to benchmark, improve, or reuse the models for educational purposes.

Overall, the project combined NLP, model optimization, fine-tuning techniques, and efficient deployment, resulting in a functional AI tutoring platform capable of adaptive, personalized learning experiences.