title	AI Employee Work Simulator
emoji	💼
colorFrom	blue
colorTo	indigo
sdk	docker
app_port	8000
pinned	false
tags	openenv

AI Employee Work Simulator

An OpenEnv environment where an AI agent acts as an employee completing real-world office tasks under deadline pressure. Built for the OpenEnv Round 1 Hackathon by team WooshiWooshi.

Environment Description

The agent must complete 3 tasks of increasing difficulty by selecting tasks, submitting formatted answers, and managing a limited step budget. Each task has a deterministic grader scoring 0.0–1.0.

Validated: 6/6 OpenEnv compliance checks pass. Live score: 3.45 / 3.45 max (100%).

Tasks

Task ID	Difficulty	Description	Max Score
email_categorizer	Easy	Classify 10 emails as urgent, normal, or spam	1.0
bug_prioritizer	Medium	Rank 5 bug reports by priority (highest → lowest) using severity + users affected	1.0
sprint_planner	Hard	Order 8 work items respecting dependencies, within a 30h budget, minimising time to T4	1.0

Action Space

{
  "action_type": "select_task | submit_answer | request_info | skip_task",
  "task_id": "<task id — only for select_task>",
  "content": "<JSON answer string — only for submit_answer>"
}

Action	Effect	Reward
select_task	Activate a task	0.0
submit_answer	Grade the answer	0.0–1.15 (score + first-attempt bonus)
request_info	Get a hint	−0.05
skip_task	Abandon task	−0.40

Observation Space

{
  "step": 3,
  "steps_remaining": 22,
  "tasks": [{"task_id": "...", "difficulty": "...", "status": "...", "score": 0.0, "attempts": 0}],
  "current_task_id": "email_categorizer",
  "last_action_result": "Task completed! Score: 1.0000.",
  "total_reward": 1.15,
  "instructions": "..."
}

Reward Function

• Dense: partial credit on every submission (score × 1.0)
• First-attempt bonus: +0.15 for scoring ≥ 0.9 on first try
• Step budget: 25 steps per episode across all 3 tasks
• Max possible: ~3.45 across all 3 tasks

Graders

Task	Method
email_categorizer	Exact match per email (e5 accepts urgent or normal)
bug_prioritizer	Kendall's tau — partial credit for near-correct orderings
sprint_planner	3-component score: dependency order (40%) + budget (30%) + T4 speed (30%)

Quick Start

from ai_employee_env import AiEmployeeAction, AiEmployeeEnv

async with AiEmployeeEnv.from_env("piroBeastie/ai-employee-env") as env:
    await env.reset()
    await env.step(AiEmployeeAction(action_type="select_task", task_id="email_categorizer"))
    result = await env.step(AiEmployeeAction(
        action_type="submit_answer",
        content='{"e1":"urgent","e2":"normal","e3":"spam","e4":"urgent","e5":"normal","e6":"spam","e7":"urgent","e8":"normal","e9":"urgent","e10":"normal"}'
    ))
    print(result.reward)  # 1.15 (score 1.0 + first-attempt bonus 0.15)

Local Setup

pip

pip install openenv-core fastapi uvicorn pydantic
uvicorn server.app:app --host 0.0.0.0 --port 8000

Docker

docker build -t ai-employee-env -f server/Dockerfile .
docker run -p 8000:8000 ai-employee-env

Test & Validate

curl http://localhost:8000/health
openenv validate --url http://localhost:8000

Run Inference

export HF_TOKEN=your_token_here
export API_BASE_URL=https://router.huggingface.co/v1
export MODEL_NAME=Qwen/Qwen2.5-72B-Instruct
python inference.py

Live Space

https://huggingface.co/spaces/piroBeastie/ai-employee-env