RN
HomeWorkExperienceAboutContact
All projectsProject 01 of 08
Internship project2026

AI Moderator

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.

  • LSL
  • Python
  • FastAPI
  • Flet
  • HTTP/JSON
  • Ollama
  • Qwen
Interface preview
Build typeInternship project
Year2026
ContributionIntegrated the in-world chat and jail objects with moderation logic, session state, AI-assisted review, and a reusable Flet monitoring interface.

Explore project

OverviewArchitectureTechnical workFeaturesGallery

Overview

Moderation with clear control

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

From chat to clear action

A traceable path from input to output, with each layer visible.

  1. 01System layer
    LSL Chat ObjectCaptures public chat in world
  2. 02System layer
    HTTP RequestSends message context as JSON
  3. 03System layer
    FastAPI BackendReceives and routes requests
  4. 04System layer
    Rule NormalizationChecks deterministic rules first
  5. 05System layer
    AI AnalysisReviews ambiguous content
  6. 06System layer
    Violation StateStores the moderation decision
  7. 07System layer
    Prison / AI JailerApplies and resolves restrictions
  8. 08System layer
    Dashboard MonitoringSurfaces live operational state

Technical work

Core technical decisions

  1. 01

    Moderation handoff: The LSL object sends avatar identity, region, position, and chat as JSON.

  2. 02

    Selective analysis: Normalization and safe or blocked terms handle deterministic cases before AI-assisted review.

  3. 03

    Session continuity: Prison state retains the context needed to return an avatar to the original location.

  4. 04

    Operational visibility: Dashboard clients validate responses and preserve empty or unavailable states without invented data.

Features

What it can do

01

Real-Time Chat Monitoring

Public chat moves from an interactive world object to the moderation backend.

02

Rule-Based + AI Evaluation

Deterministic normalization precedes AI-assisted analysis when needed.

03

Automated Prison Workflow

Violation state and location context coordinate restriction and return.

04

AI Jailer Interaction

Apology messages can be reviewed before a release action is returned.

05

Monitoring Dashboard

Overview, analytics, prisoners, violations, and settings expose system state.

06

Reusable UI Components

Cards, filters, tables, navigation, empty states, and badges are separated by role.

Gallery

Screens and scenes

Swipe through the screens and scenes. Open a frame for details.

Learning through what i build

© 2026 Rafa'Na'ilah Septia