Outcome-Based Education (OBE) is structured around defined, measurable learning outcomes aligned with academic curricula and professional competencies. While OBE emphasizes learning goals, traditional evaluation often relies on quantitative metrics—such as scores or pass/fail outcomes—overlooking insights in structured and unstructured data. This research proposes an intelligent framework for OBE analysis using the TinyLLaMA-1.1B-Chat model, a quantized, resource-efficient model from Meta AI’s LLaMA family. Optimized for low-compute environments, it applies Natural Language Understanding (NLU) to interpret academic records. The framework includes data collection, preprocessing, model fine-tuning via Low-Rank Adaptation (LoRA), and visualization. By freezing base layers and adapting new ones, LoRA enables efficient fine-tuning. The system identifies Blooms levels, classifies question difficulty, predicts CLO/PI alignments, and summarizes feedback. Visualizations such as bar plots and heatmaps support analysis. By transforming semi-structured data and feedback into actionable insights, the model improves assessment and instructional feedback. Experiments show gains in F1 score (0.86) and CLO prediction (89%), demonstrating that lightweight models like TinyLLaMA support scalable OBE evaluation.