What is multimodal-interview-system?

This is a prototype interview system built with Next.js and a workflow automation platform called Coze. The README is written in Chinese, so this explanation is drawn from that source. The system is aimed at use cases like academic course projects, graduation theses, and research experiments involving AI-assisted interviewing. The core idea is that you paste in a candidate's resume text and a job description, and the system reads both to figure out which of seven predefined job categories applies (developer, data analyst, product manager, sales, operations, customer service, or HR). It then generates a set of interview questions along with evaluation criteria for each question. The candidate answers each question by recording a video in the browser using their camera and microphone. Once all answers are submitted, the system evaluates each response by analyzing three channels at once: the spoken text, the audio tone, and the video. After all questions are scored, it generates a final report that rolls up the individual results. This three-channel evaluation approach is what makes it "multimodal." The system runs in a mock mode by default, meaning you can go through the full product flow without connecting to any external AI service. If you want real AI-powered evaluation, you configure credentials for the Coze workflow API. The project ships with exported Coze workflow files and job templates archived in its docs folder, so others can reproduce the setup. It is built with Next.js, React, TypeScript, and uses either PostgreSQL or a local JSON file for storing interview sessions. A default admin login is provided for local development. The README emphasizes this is a prototype and research tool, not a production product.