RL-Recruiters
We built an OpenEnv RL environment that trains an LLM to act as a Staffing Agency CEO over a 52-week simulated year.
Core trick
Hidden candidate skills are scored by a background LLM judge only after an interview action, over 52 weekly steps with bench-cost penalties.
Description
We built an OpenEnv RL environment that trains an LLM to act as a Staffing Agency CEO over a 52-week simulated year. Inspired by VendingBench, it features brutal economics: hired candidates sitting on the "bench" bleed weekly salaries, while massive profits only unlock when multi-role corporate projects are fully staffed before strict deadlines. The agent starts blind. It must spend actions to "interview" candidates, triggering a background LLM Judge to score hidden skills and flag culture risks.
The Problem It Solves:
1. Long-Horizon Planning (Statement 2): We push LLMs beyond shallow, single-turn reasoning. The agent must survive sparse, delayed rewards and manage a complex P&L over 52 steps, avoiding the trap of blindly hiring everyone.
2. HR Workflow (Scale AI): Traditional staffing agencies are bloated and slow. We decentralize human capital routing so every solo freelancer can operate with the capacity of a million-dollar agency.
README excerpt
--- title: Staffing Agency emoji: 🏢 colorFrom: blue colorTo: purple sdk: docker app_port: 7860 pinned: false --- # StaffingGym: Teaching an LLM to Run a Staffing Agency with Reinforcement Learning > **OpenEnv Hackathon — March 2026** · Multi-Agent Interactions + Long-Horizon Planning (Scale AI / Mercor sub-theme) > > An LLM agent acts as a **Staffing Agency CEO** managing multiple clients and candidates > over a 52-week simulated business year. The agent must balance bench costs, candidate > patience, project deadlines, and client satisfaction to maximise profit using real > tool calls against a live environment server — no fabricated rewards. --- ## Why We Built This This project didn't start at a hackathon. It started with frustration. We've worked with staffing agencies to find the right talent for our teams. The process is painful — weeks of back-and-forth, mismatched candidates, lost time on both sides. The agency bleeds money keeping developers on the bench. The company bleeds money waiting for the right hire. Everyone loses. The insight that wouldn't leave us alone: **what if one person with a laptop could spin up a staffing agency?** Not a job board. Not a matching algorithm. A full autonomous agent that sources, interviews, hires, and places developers into multi-role corporate projects — managing a real P&L the entire time. We're building **RL Recruiters** — an autonomous RL-based environment inspired by the [VendingBench](https://arxiv.org/abs/2410.18572) m