@yaojingang: The daily recruitment and interview volume for the team is growing rapidly. So in the past couple of days, we rolled out a major upgrade to our "AI Interview Butler" system. The core is to achieve more automated and intelligent four-terminal collaboration: AI Interview System + GitHub + Feishu + Codex. 1. AI Interview Butler Main System: Responsible for candidates, resumes, automated AI initial interviews, automated interview reports, feedback, scheduling, data analysis reports, sync status, and other core business data—essentially the business hub of the entire recruitment process.

The daily recruitment and interview volume for the team is growing rapidly. So in the past couple of days, we rolled out a major upgrade to our "AI Interview Butler" system. The core is to achieve more automated and intelligent four-terminal collaboration: AI Interview System + GitHub + Feishu + Codex. 1. AI Interview Butler Main System Responsible for core business data such as candidates, resumes, automated AI initial interviews, automated interview reports, feedback, scheduling, data analysis reports, and sync status—acting as the business hub of the entire recruitment process. 2. GitHub Repository GitHub repository serves as the collaboration space for the recruitment process. Each candidate has their own data directory, and the process stage is automatically recorded and transitioned through directories, workflow.json, candidate-index.csv, and Git commits. This means that a candidate's transition journey naturally forms a versioned chain of evidence. 3. Codex Skill Acts as the local intelligent operation layer. Each interviewer actively pulls the candidate's data package, generates evaluations, writes initial/second interview feedback, saves scheduling information, submits Feishu tasks, interview feedback, and data analysis, etc. 4. Feishu Execution Layer Through the Feishu CLI, it automatically creates interview calendar events, sends private messages and reminder prompts. A background worker handles async tasks, failure retries, and status write-backs, including weekly interview data report sync. The overall flow is roughly as follows: 1. HR or candidate uploads a resume, and the system asynchronously parses structured information. 2. After the candidate completes the AI initial interview, the system saves Q&A, conversation history, keywords, and scores. 3. The report generation service asynchronously produces HR and candidate versions of the report, and the system agent automatically syncs them to the GitHub repository. 4. Colleagues can view candidate information and related materials in GitHub or Codex, generate evaluations based on the data package, and write feedback. 5. If the feedback result is pass, the candidate automatically moves to the next stage directory in the GitHub repository; if reject or hold, only the status and feedback record are updated, and the directory is not moved. 6. Once the interview schedule is confirmed, the system creates a Feishu calendar task, the worker executes it asynchronously, and sends feedback reminder prompts 1 hour and 3 hours after the interview. For interviewers, the core collaborative and interactive interface becomes Codex, while all other terminals automatically collaborate and transition. In this architecture, the most interesting aspects are: 1. The database stores primary business data; GitHub stores the evidence chain of collaboration. 2. The recruitment process is GitOps-ified: stage changes, feedback records, and directory moves are all traceable. 3. Local agents actively pull data, enabling intelligent collaboration while controlling the security boundary. 4. All Feishu tasks are executed asynchronously, allowing fault isolation between calendar, notifications, and reminders. 5. AI handles fact extraction and suggestions; humans handle conclusions, reasons, and responsibility judgments. 6. Candidates are no longer just a row in a backend table, but a data object that can be carried, versioned, written back to, and reached. In this system, candidates are remodeled as a state object that spans the AI interview, GitHub collaboration, Codex local agent, and Feishu reachability. This gives the recruitment process four capabilities simultaneously: automation efficiency, human responsibility boundaries, organizational recruitment process memory, and operational traceability.