Real-Time Chat Monitoring
Public chat moves from an interactive world object to the moderation backend.
AI-Assisted Chat Moderation for Virtual Worlds
Built an integrated moderation system connecting interactive LSL objects, a FastAPI backend, AI-assisted analysis, an automated prison workflow, and a monitoring dashboard.
Overview
An internship project connecting public chat from a virtual world to a backend moderation workflow with clear operational states in the interface.
Public chat is sent by the LSL moderation object to FastAPI. The backend normalizes the message, checks safe and blocked terms, and requests AI analysis when required. A violation creates a prison session; the jailer can process an apology and release the avatar to the stored location. Dashboard endpoints expose the current state for monitoring.
Architecture
A traceable path from input to output, with each layer visible.
Technical work
Moderation handoff: The LSL object sends avatar identity, region, position, and chat as JSON.
Selective analysis: Normalization and safe or blocked terms handle deterministic cases before AI-assisted review.
Session continuity: Prison state retains the context needed to return an avatar to the original location.
Operational visibility: Dashboard clients validate responses and preserve empty or unavailable states without invented data.
Features
Public chat moves from an interactive world object to the moderation backend.
Deterministic normalization precedes AI-assisted analysis when needed.
Violation state and location context coordinate restriction and return.
Apology messages can be reviewed before a release action is returned.
Overview, analytics, prisoners, violations, and settings expose system state.
Cards, filters, tables, navigation, empty states, and badges are separated by role.
Gallery
Swipe through the screens and scenes. Open a frame for details.