Builder · Public · OpenEnv Hackathon India 2026
PHANTOM
PHANTOM is a training environment for cognitive security. A defender agent handles a live intrusion while an adversary agent feeds it made-up SIEM log entries.
How it trains
Both agents learn against each other with multi-agent reinforcement learning. I used GRPO group sampling, a curriculum that adapts as the agents improve, and an LLM to judge the results. I built it for the Meta × PyTorch × Hugging Face OpenEnv Hackathon India 2026.
- LinkCode ↗