Spaces:
Configuration error
Download README.md from MetaStableMinds/business-sim-openenv: direct link, hf CLI and curl.
- Browser
- Download file 4.62 kB
-
https://huggingface.co/spaces/MetaStableMinds/business-sim-openenv/resolve/main/README.md
- Command line
-
hf download hf://spaces/MetaStableMinds/business-sim-openenv/README.md
-
curl -L -o README.md https://huggingface.co/spaces/MetaStableMinds/business-sim-openenv/resolve/main/README.md
๐ข AI CEO: Strategic Business Simulation Agent
Meta PyTorch Hackathon โ OpenEnv Round 1 Submission
๐ Overview
This repository contains a state-of-the-art AI CEO Agent developed for the Meta PyTorch Hackathon. The agent is designed to manage a software company through a highly complex, adversarial, and stochastic business simulation environment.
๐ง Problem Statement
The BusinessSimEnv is a multi-turn, delayed-reward environment where every action has cascading consequences. A CEO must survive up to 8 quarters (2 years) while maintaining:
- Positive cash flow in a recession.
- Low employee burnout to avoid skill degradation.
- High reputation to attract premium projects.
The AdversarialAgent in the environment injects realistic "shocks" (budget audits, key departures, and client disputes) to test the agent's resilience.
๐ The Simulation Ecosystem: A Tri-Agent Interaction
Our simulation is not a static set of rules, but a dynamic, multi-agent world where the AI CEO must compete and survive against two other environmental agents:
๐ค 1. The AI CEO (The Decision Engine)
- Role: The primary protagonist of the simulation.
- Actions: Project acceptance, hiring/firing (Salary Optimization), tech stack selection, and skill training.
- Objective: Maximize long-term corporate health (Budget, Reputation, and Low Burnout) through high-reasoning ROI analysis.
๐ 2. The Market Agent (The Economic Engine)
- Role: Simulates the macro-economic environment and consumer demand.
- Behavior: Driven by a Markov Chain, this agent triggers Boom, Stable, and Recession cycles.
- Impact: It alters the Domain Demand multipliers (e.g., higher demand for AI during Booms) and profit variances, forcing the CEO to adapt their strategy quarterly.
๐ก๏ธ 3. The Adversarial Agent (The Friction Engine)
- Role: Simulates the "Chaos" of real-world business operations.
- Impact: Triggers events like Budget Audits, Key Employee Departures, and Client Disputes.
๐ ๏ธ Strategic Architecture
Our agent implements a Reasoning-Driven Control Loop with the following advanced heuristics:
The agent leverages GPT-OSS 20B to make quarterly decisions across six crucial business factors:
- Profit Potential: Strategic project selection.
- Risk Management: Identifying and mitigating "Hidden Risk."
- Team Capability: Hiring, firing, and training management.
- Resource Optimization: Budgeting and tool choice (Tech Stack).
- Market Alignment: Adapting to Boom/Recession cycles.
- Reputation: Building long-term brand equity vs. short-term gains.
๐ 1. Proactive ROI Analysis
The agent performs real-time Return on Investment (ROI) analysis factoring in Expected Profit, Total Risk, and Quarterly Salary Burn.
๐ 2. Salary Optimization Hack
The agent dynamically fires developers to instantly save ~$24,000 per quarter in overhead.
๐ก๏ธ 3. Bankruptcy Protection
Follows a strict Budget-to-Penalty Ratio. Skip the quarter rather than taking a "suicidal" gamble.
๐ Evaluation Methodology
Our agent is validated against three distinct "Gates":
- Unit Tests: Verifying deterministic logic for budget calculations (
test_grade.py). - Adversarial Stress Test: Running the Hard task (
adversarial_resilience). - Generalization: Heuristics that work across all task difficulties.
๐ ๏ธ Tech Stack & Dependencies
- Core Logic: Python 3.12+ with Pydantic schemas.
- Inference: GPT-OSS 20B on Hugging Face Router.
- Environment API: FastAPI simulation server.
- Communication: Asynchronous Httpx.
๐ฎ Future Roadmap (v2.0)
- Multi-Agent Negotiation: Direct negotiation with the Adversarial Agent.
- Dynamic Tech Stacks: Modular technology selection.
- Reinforcement Learning (PPO): Transitioning from heuristic prompts to fine-tuned models.
โ๏ธ Installation & Usage
- Configure Token:
$env:HF_TOKEN="your_token" - Run Inference:
python inference.py - Validate:
python validate.py