Selected AI Projects

01 BiasInspector LLM Agents · Bias Detection 02 SituationalPriv Multimodal Privacy Detection

BiasInspector

LLM Multi-Agent Framework for Structured Data Bias Detection

LLM Agents Multi-Agent Systems RAG LangChain Bias Detection Benchmark

Built an end-to-end multi-agent framework for structured-data bias detection, integrating RAG-based method retrieval, statistical analysis, visualization, and report generation within a unified automated pipeline.

System Overview

Implemented the agent workflow with LangGraph and LangChain, enabling agent orchestration, tool calling, stateful task execution, and multi-step reasoning for complex bias-analysis tasks.

Agent Workflow

46 analytical tools

100+ retrievable methods

100 benchmark tasks

SituationalPriv

Context-Aware Multimodal Privacy Detection System

GitHub Paper

Multimodal AI Vision-Language Models Privacy Detection GPT-4o LLaVA Grounded-SAM

Built a context-aware multimodal system for privacy-risk assessment, enabling automated detection of sensitive visual information across different real-world sharing scenarios.

Multimodal Privacy Detection Workflow

Developed an LLM-guided multimodal workflow integrating GPT-4o, LLaVA, and RAM-Grounded-SAM for visual grounding, context-aware questioning, and privacy-risk reasoning.

440 privacy-related images

LLM + VLM collaborative framework

NeurIPS 2025 workshop publication

Haoxuan Li