This WPI IQP project involved developing an AI-powered writing assistant for TOEFL/IELTS preparation, working with a Chinese educational company (comparable to Google Classroom) to design and implement a prototype system for evaluating student writing. The core idea combines semantic-based embedding and algorithmic-based vectorization with retrieval-augmented generation (RAG) to assist a large language model in evaluating essays. The system ingests an essay, generates two distinct performance vectors (one LLM-based and embedded, another algorithmically derived using error counters), then retrieves relevant exemplar essays using the RAG before generating scored feedback.

I single-handedly designed and implemented the writing evaluation portion of the system, including the RAG architecture, the document ingestion pipeline, database management, and API integration. The system scores essays across four TOEFL dimensions: task response, coherence and cohesion, lexical resource, and grammatical range and accuracy. The full project was published in a WPI group report, linked below.

Read the full published report: digital.wpi.edu