AI Interview Assistant

The problem

Research by Ji et al. (2023) highlights that Large Language Models can generate fluent but unsupported responses when relying only on their internal parameters, commonly known as the hallucination problem. In interview preparation, a general-purpose AI assistant lacks candidate-specific context, which can lead to questions and feedback that are not personalized to the candidate's resume, projects, and experience.

The solution

I implemented a RAG pipeline that processes resumes through parsing, semantic chunking, embedding generation, vector similarity search, and context-based prompting. Retrieved resume information is provided to the LLM to generate more relevant interview questions and evaluations.