AI简历筛选与面试验证系统

Applicant Screening and Interview Validation System

An AI-powered applicant screening and interview validation system using open-source LLMs via Ollama, Groq, or OpenAI.

Architecture

The system is a three-stage hiring pipeline exposed as a Flask REST API. A resume submitted to /screen is parsed for contact details and skills, then handed to a ScreeningAgent that asks an LLM to score and evaluate it against job requirements. The result is stored in memory and an applicant ID is returned.

When interview responses arrive at /validate, two things run: a ValidationAgent calls the LLM for a semantic evaluation across technical ability, communication, and problem-solving, while a separate MCP Server process (running over SSE on :8000) applies rule-based checks — minimum question count, response length, and keyword scoring. The final interview score blends both: 60% LLM, 40% MCP rules.

Calling /recommend/<id> pulls both evaluations and sends them together to the LLM in a single prompt, which returns a STRONG_HIRE / HIRE / MAYBE / REJECT decision with a confidence score and reasoning.

The LLM layer is provider-agnostic — a single LLM_PROVIDER environment variable switches between Ollama (local, free), Groq (cloud, free), and OpenAI without changing any application code. All state lives in plain Python dicts in memory; there is no database.

Setup

1. Install dependencies

pip install -r requirements.txt

2. Choose an LLM provider

Copy .env.example to .env and set your provider:

cp .env.example .env

Option A: Ollama (free, local — recommended)

Install Ollama from https://ollama.com, then pull a model:

ollama pull llama3.1

Set in .env:

LLM_PROVIDER=ollama
LLM_MODEL=llama3.1
OLLAMA_BASE_URL=http://localhost:11434

Option B: Groq (free cloud API)

Get a free API key at https://console.groq.com.

Set in .env:

LLM_PROVIDER=groq
LLM_MODEL=llama-3.1-8b-instant
GROQ_API_KEY=your_key_here

Option C: OpenAI

LLM_PROVIDER=openai
LLM_MODEL=gpt-4o-mini
OPENAI_API_KEY=your_key_here

3. Run the application

python src/app.py

The API will start at http://localhost:5000.

4. (Optional) Start the MCP interview validator server

python src/mcp_servers/interview_validator.py

API Endpoints

Example: Screen a resume

curl -X POST http://localhost:5000/screen \
  -H "Content-Type: application/json" \
  -d '{
    "resume_text": "Jane Doe\njane@example.com\nSoftware Engineer with 5 years Python, AWS, Docker experience.",
    "job_requirements": {
      "required_skills": ["Python", "AWS"],
      "required_experience": "3+ years"
    }
  }'

Example: Validate an interview

curl -X POST http://localhost:5000/validate \
  -H "Content-Type: application/json" \
  -d '{
    "applicant_id": 1,
    "interview_data": {
      "responses": {
        "Q1": "I have 5 years of Python experience building microservices.",
        "Q2": "I led a team of 4 engineers on a cloud migration project.",
        "Q3": "I use TDD and CI/CD to ensure code quality."
      }
    }
  }'