Leveraging LLM and RAG for Automated Answer Script Evaluation

In the realm of education, examinations play a very significant role. If writing an exam is considered tedious, the process of evaluating hundreds of answer scripts can be even more daunting. It usually takes weeks to evaluate all the answer scripts, and there is always a factor of bias in human corrections. Hence, to eliminate all these problems, this study uses a large language model to automate this task and eliminate human bias. The study leverages a fine-tuned LLM to assess answer scripts related to the subject of operating systems using a specific dataset. Along with the LLM, it incorporates Retrieval-Augmented Generation (RAG) to get the context of a given question from a prescribed textbook. The proposed platform also has the capability of analyzing handwriting from an actual answer script, and then this is passed onto the model as input along with context from RAG. Finally, the entire system is integrated into an interactive web platform, deployed using AWS SageMaker. By combining all the technology mentioned, this study has made a sincere attempt to solve the burden of correcting manuscripts.