Google SWE deploys 48 AI agents to discover Maya pyramid using NASA Lidar data

The intersection of high-performance computing and archaeological discovery reached a remarkable milestone when a Google Software Engineer successfully directed 48 autonomous AI agents to analyze massive NASA Lidar datasets. Within a mere 72 hours of continuous execution, the multi-agent system successfully mapped and identified an entirely unmapped Maya pyramid. This achievement marks a massive shift from traditional archaeological surveying, which typically takes decades of manual field expeditions and topographical analysis. For developers and enterprises looking to replicate this level of high-throughput automated data processing, understanding how to orchestrate multi-agent LLM workflows efficiently is now an essential engineering skill.

The Architecture Behind the 48-Agent Swarm

Building a resilient multi-agent system that can ingest petabyte-scale geographic data requires a robust backend architecture. Unlike standard single-prompt chatbot implementations, this archaeological project relied on a hierarchical agentic framework. The pipeline was broken down into specialized operational tiers: data ingestion, spatial anomaly detection, topographical classification, and cross-referencing historical cartography.

To handle the immense computational load and parallel API calls without hitting rate limits or incurring prohibitive costs, developers frequently turn to agile infrastructure solutions. This is where leveraging a high-performance LLM aggregator like n1n.ai becomes critical. By routing requests across multiple frontier models—such as Claude 3.5 Sonnet for complex spatial reasoning and DeepSeek-V3 for rapid pattern matching—the agent swarm maintained optimal throughput and cost efficiency throughout the 72-hour run.

Setting Up Your Own Multi-Agent Pipeline with Python

To help developers implement similar high-speed data analysis workflows, let us look at a practical Python implementation using an asynchronous agent loop. Below is a streamlined example demonstrating how to query multiple LLM endpoints concurrently using a unified API aggregator.

import asyncio
import aiohttp
import json

API_URL = "https://api.n1n.ai/v1/chat/completions"
API_KEY = "your-n1n-ai-api-key"

async def query_agent(session, agent_id, spatial_data_chunk):
    headers = \{
        "Authorization": f"Bearer \{API_KEY\}