Interview Voice Agent · 面试语音助手

Resume-grounded, bilingual (EN + 中文) mock interview copilot with browser voice input and streaming responses.

Quickstart · Features · Architecture · API · Notes

---

What this is

This project helps you practice mock interviews by generating interview-ready answers that are grounded in your resume.json, while supporting text + voice input and streaming output in the browser.

Features

• 🎤 Voice + Text: Web Speech API recognition in the browser (Chrome recommended)
• 🧾 Resume-grounded: answers are anchored to resume.json to avoid made-up experience
• ⚡ Streaming: token streaming from FastAPI to UI
• 🧠 Session memory: /new-session + /chat/stream
• 🌍 Bilingual output: English answer + Chinese explanation (固定输出结构)

Demo (local)

• Start the server, then open http://127.0.0.1:8000
• Type a question, or click the voice button to speak

Architecture

flowchart TD
  UI["Browser UI<br/>Text + Voice"] -->|"Web Speech API"| ASR["Speech Recognition"]
  UI -->|"HTTP"| API["FastAPI"]
  API -->|"load"| RESUME["resume.json"]
  API -->|"prompt + resume grounding"| LLM["OpenAI Chat Completions<br/>streaming"]
  LLM -->|"tokens"| API
  API -->|"StreamingResponse"| UI

Loading

Quickstart

Prerequisites

• Python 3.10+
• An OpenAI API key in OPENAI_API_KEY

Install

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirement.txt

Configure API key

export OPENAI_API_KEY="your_api_key"

Prepare resume.json

This file is loaded on startup and injected into the system prompt.

Example:

{
  "name": "Your Name",
  "skills": ["Java", "Kubernetes"],
  "projects": [
    {
      "name": "Project A",
      "tech_stack": ["Spring Boot"],
      "details": ["Built APIs"],
      "impact": "Improved performance"
    }
  ]
}

Run

uvicorn main:app --reload

Then open http://127.0.0.1:8000.

How it works (files)

• main.py: loads resume.json, builds a bilingual system prompt, streams model tokens back to the browser
• static/index.html: minimal chat UI + voice recognition + optional TTS

API

• GET /new-session → returns { "session_id": "..." }
• POST /chat/stream → streams plain text tokens

Request body:

{ "session_id": "...", "message": "..." }

Output format (model)

Click to expand

[Keywords]
- ...

[Answer - English]
...

[Answer - 中文]
...

Notes & Troubleshooting

• Voice not working: use Chrome, allow microphone permissions, then reload.
• Mixed Chinese/English recognition: tweak recog.lang in static/index.html (e.g. zh-CN / en-US).
• Prompt too long / slow: keep resume.json concise (avoid large raw text blocks).
• Missing API key: ensure OPENAI_API_KEY is set in your shell environment.

Roadmap

• Real-time voice streaming (no manual stop)
• Better UI/UX (chat bubbles, markdown rendering)
• Persistent chat history (Redis/DB)
• Multi-user auth + interviewer mode