The Problem
High-volume event and mobility submissions need fast, accurate risk screening — with enough transparency for reviewers to trust automated flags.
What I Built
A multi-agent AI risk-classification system that combines classical ML signals with LLM reasoning to analyze event, mobility, and persona data and flag high-risk cases.
How It Works
Specialized agents handle data extraction, risk scoring, and validation. LangChain orchestrates the pipeline while Langfuse traces every agent run for auditability. OpenRouter provides model routing across the classification stages.
Outcome / Results
Ranked #64 out of 1,920 international teams in the Reply AI Agent Challenge 2026 — validating the hybrid ML + LLM approach under real competition pressure.