A Mathematical Framework for Custom Reward Functions in Job Application Evaluation using Reinforcement Learning

TL;DR

This work proposes a two-stage fine-tuning framework to improve resume evaluation beyond keyword-based ATS by using a compact small language model. It first applies Supervised Fine-Tuning (SFT) on recruitment-related data, then refines the policy with Generative Reward Policy Optimization (GRPO) guided by a multi-criteria reward to align with recruiter judgment while mitigating reward hacking. The reward function is a weighted sum across four criteria, optimized with KL regularization to maintain alignment with the SFT baseline. Empirical results on synthetic data show the GRPO phase yields higher accuracy (91.4%) and better SELECTED-class F1 (0.92) than SFT alone, with zero false positives in the test set, demonstrating the viability of a gentle, two-phase approach for domain-specific, human-aligned candidate scoring.

Abstract

Conventional Applicant Tracking Systems (ATS) tend to be inflexible keyword-matchers, and deny gifted candidates a role due to a few minor semantic mismatches. This article describes a new two-step process to design a more refined resume evaluation model based on a small language model (<600M parameters) that is finetuned using GRPO on a custom reward function. To begin with, Supervised Fine-Tuning (SFT) was used to build a solid baseline model. Second, this SFT model was also optimized with the help of Reinforcement Learning (RL) through GRPO under the guidance of a new, multi-component reward function that can holistically assess candidates beyond simple keyword matching. We indicate that the RL application presents a critical problem of reward hacking due to the initial experiments of aggressive penalties, which produces faulty, excessively negative model behaviors. We have overcome this challenge by refining the reward function repeatedly and training hyperparameters into a stable "gentle polishing process" of the reward function. Our resulting GRPO-polished model demonstrates significant real-world efficacy, achieving a final accuracy of 91% on unseen test data. The model shows a strong ability to correctly identify qualified candidates (recall of 0.85 for the 'SELECTED' class) while also showing exceptional precision (1.0), confirming its reliability. These results indicate that a properly executed, two-step fine-tuning procedure can indeed effectively refine a small language model to be able to conduct fine-tuned and human-like candidate scoring, overcoming the drawbacks of both traditional ATS and naive RL usage.

Figures (4)

• Figure 1: Training and Validation Loss Curves for SFT
• Figure 2: GRPO Training Loss over 337 steps
• Figure 3: Bar chart comparing SFT-only and GRPO-refined model performance
• Figure 4: Confusion matrices for SFT and GRPO models