Harvey: An Intelligent Recruitment Assistant Based on Hybrid Multi-Model Architecture

Harvey is an Agentic HR intelligent assistant developed by Nathan Mendis and released on GitHub on June 12, 2026. Focused on recruitment scenarios, it uses a hybrid multi-model architecture to balance response speed and professional reasoning capabilities. It aims to automate the entire recruitment process, including resume screening, interview assistance, and process management, to improve efficiency and quality, providing a reference for AI applications in vertical domains.

Project Origin

• Original author/maintainer: Nathan Mendis
• Source platform: GitHub
• Release date: June 12, 2026

Project Overview

Harvey is an important exploration of Agentic AI in the enterprise HR recruitment field. Unlike general-purpose AI assistants, it deeply focuses on recruitment scenarios and achieves a balance between professional capabilities and response speed through a hybrid multi-model architecture.

Harvey's core innovation lies in its hybrid multi-model architecture, combining models with different strengths for collaboration:

Lightweight Intent Detection Model

Quickly identifies the type of user input intent (5 types of HR tasks including resume screening, interview scheduling, candidate evaluation, etc.), responds in real time, and routes complex tasks to appropriate modules.

Deep Reasoning Model

Handles complex analysis tasks:

• Resume parsing and structuring
• Skill matching analysis
• Interview question generation
• Comprehensive evaluation report generation

Harvey includes three core functional modules:

Automated Resume Screening

Batch parses resumes to extract key information, calculates matching scores based on job descriptions, identifies outstanding candidates, and detects highlights and risk points.

Intelligent Interview Assistance

Generates personalized interview questions, records interview key points in real time, produces structured evaluation reports after interviews, and supports comparison of information from multiple interview rounds.

Recruitment Process Management

Proactively advances the process: tracks candidate progress, automatically sends notifications, coordinates interviewers' schedules, and generates recruitment funnel analysis reports.

Domain Knowledge Integration

Integrates industry terminology and skill ontologies, understands professional requirements for different positions, and supports parsing of multiple resume formats.

Enterprise-Level Security Design

Emphasizes candidate data privacy protection, access permission control, audit log recording, and compliance checks.

Efficiency Improvement

Resume initial screening time is reduced from hours to minutes; interview scheduling and evaluation report generation are automated.

Quality Assurance

Reduces subjective bias in manual screening, ensures comprehensiveness of evaluation dimensions, and retains complete decision-making basis and process records.

Scalability

The hybrid architecture adapts to enterprises of different sizes: small teams use lightweight configurations to control costs, large enterprises expand reasoning capabilities to handle complex scenarios, and multi-language and multi-region requirements are supported.

Harvey demonstrates the application paradigm of Agentic AI in vertical domains:

1. Domain specialization is better than general generalization
2. Hybrid model architecture balances cost and performance
3. Human-machine collaboration enhances human capabilities (allowing HR to focus on high-value judgment and decision-making)

Recommendation for developers: This architecture provides a valuable reference for the application of LLMs in enterprise scenarios.