Spaces:
Sleeping
Sleeping
| name: synth_audit_env | |
| title: "SynthAudit.Env β Multi-Agent Clinical AI Oversight" | |
| description: > | |
| A multi-agent OpenEnv environment for training oversight agents | |
| to monitor, audit, and correct medical AI decisions. The Actor | |
| agent proposes clinical diagnoses; the Oversight agent catches | |
| errors, hallucinations, and bias blind spots using SHAP analysis. | |
| version: "1.0.0" | |
| theme: "Multi-Agent Interactions β Fleet AI: Scalable Oversight" | |
| author: "Sumit Saraswat" | |
| server: | |
| dockerfile: server/Dockerfile | |
| port: 8000 | |
| models: | |
| action: models.SynthAuditAction | |
| observation: models.SynthAuditObservation | |
| state: models.SynthAuditState | |
| tasks: | |
| oversight_easy: | |
| description: "Easy oversight β catch obvious age violations" | |
| difficulty: easy | |
| max_steps: 25 | |
| oversight_medium: | |
| description: "Medium oversight β catch age, temporal, and window errors" | |
| difficulty: medium | |
| max_steps: 40 | |
| oversight_hard: | |
| description: "Hard oversight β catch subtle comorbidity overrides and bias" | |
| difficulty: hard | |
| max_steps: 55 | |