
InsightAgent is a multi-agent AI business intelligence platform that makes AI decision-making transparent and interactive. Unlike traditional black-box AI tools, InsightAgent lets users watch multiple specialized AI agents collaborate in real-time on AMD Instinct MI300X hardware. The platform features a 4-agent pipeline: • Parser Agent — Extracts and structures data from uploaded CSV files, detecting column types and computing statistics • Vision Agent — Powered by Qwen2.5-VL-7B-Instruct for visual data analysis and chart recommendations • Analyst Agent — Uses Qwen3-8B to perform statistical analysis, identify trends, detect anomalies, and find correlations • Reporter Agent — Synthesizes all findings into executive reports with key findings and actionable recommendations What makes InsightAgent unique is the live orchestration visibility. Every agent action is streamed to the frontend via WebSocket, showing real-time progress bars, activity logs, and status transitions. Users don't just get results — they understand HOW the AI reached its conclusions. After the pipeline completes, users can Chat with their Data — asking natural language questions like "Which region generates the most revenue and why?" and receiving detailed, data-grounded answers powered by live Qwen3-8B inference on the MI300X GPU. Response latency and token metrics are displayed for full transparency. The tech stack leverages AMD's ecosystem end-to-end: vLLM v0.17.1 with ROCm serves Qwen models on the MI300X's 192GB HBM3 memory, enabling high-throughput inference. The frontend is built with React 19 and features a premium dark-mode UI with interactive charts, sortable data tables, and KPI dashboards. InsightAgent demonstrates how AMD's GPU infrastructure can power sophisticated multi-agent AI workflows with real-time user interaction — making enterprise-grade business intelligence accessible and transparent.
10 May 2026