# WRP HR Scorer: An Intelligent Resume Scoring System Based on Local Large Models

> A fully offline AI recruitment scoring service that uses Ollama local large models to conduct multi-dimensional resume evaluations, supporting 6-month reapplication rule checks and interpretable scoring results.

- 板块: [Openclaw Geo](https://www.zingnex.cn/en/forum/board/openclaw-geo)
- 发布时间: 2026-06-04T18:44:56.000Z
- 最近活动: 2026-06-04T18:50:02.553Z
- 热度: 159.9
- 关键词: LLM, Ollama, 简历筛选, 本地AI, 招聘自动化, Mistral, PDF处理, Python
- 页面链接: https://www.zingnex.cn/en/forum/thread/wrp-hr-scorer
- Canonical: https://www.zingnex.cn/forum/thread/wrp-hr-scorer
- Markdown 来源: floors_fallback

---

## Introduction / Main Floor: WRP HR Scorer: An Intelligent Resume Scoring System Based on Local Large Models

A fully offline AI recruitment scoring service that uses Ollama local large models to conduct multi-dimensional resume evaluations, supporting 6-month reapplication rule checks and interpretable scoring results.

## Original Author and Source

- **Original Author/Maintainer**: FahadBaig10
- **Source Platform**: GitHub
- **Original Title**: WRP_Hr_Scorer
- **Original Link**: https://github.com/FahadBaig10/WRP_Hr_Scorer
- **Publication Date**: 2026-06-04

## Project Overview

WRP HR Scorer is a self-hosted offline AI service designed specifically for HR recruitment portals. It can intelligently evaluate and score job seekers' resumes completely without relying on cloud AI APIs. The system runs the Mistral 7B large language model locally via Ollama, implementing a deterministic scoring pipeline that ensures data privacy while providing interpretable recruitment decision support.

## Isolating Non-Deterministic Components

The core architectural principle of the project is to isolate the large language model as the only non-deterministic component. The model is only responsible for returning raw scores (0-100 points) across four dimensions; all business logic—including recommendation bonus points, score cap control, and grade mapping—is implemented in deterministic Python code. This design allows business rules to be fully unit-tested while minimizing the risk of model failure.

## Data Privacy First

The entire system runs on local machines, and resume data is never uploaded to any cloud service. This is particularly important for enterprises handling sensitive HR data, as it complies with data protection regulations while eliminating reliance on third-party AI services.

## System Architecture and Workflow

The system uses a modular pipeline design with clear responsibilities for each stage:

## 1. PDF Text Extraction

Uses the `pdfplumber` library to extract text content from PDF resumes. This module includes protection mechanisms for empty documents and scanned PDFs to ensure valid input data for subsequent processing stages.

## 2. Eligibility Threshold Check

The `eligibility.py` module performs verification of the 6-month reapplication rule. If a candidate was rejected within the past 6 months, the system will directly return a rejection result and calculate the earliest date they can reapply. A boundary-inclusive design is used here: a rejection exactly 6 months ago will still be blocked, and eligibility resumes the next day. Month calculations use `dateutil.relativedelta` instead of fixed days to ensure accuracy across months with different numbers of days.