用AI Agent + 亚马逊实时数据API打破大卖家数据垄断：架构设计与完整实现

Tags: Amazon API AI Agent LangChain Python 电商数据 实时数据
难度： 中级 | 阅读时长： 15分钟

背景与问题

亚马逊大卖家（年GMV 1000万+）的核心竞争优势之一是实时数据能力：每15-30分钟采样竞品BSR、价格、库存，通过内部BI系统驱动实时决策。这套基础设施的自建成本每年在$60,000-$120,000之间，对中小团队完全不可及。

本文介绍一套基于 Pangolinfo Scrape API + LangChain AI Agent 的替代架构，实现相同数据能力，月成本控制在$50以内。

系统架构

┌─────────────────────────────────────────────────────┐
│                   AI Agent Layer                     │
│  LangChain Agent ← Tools: [BSR监控, 专利扫描, 报告]   │
└──────────────────┬──────────────────────────────────┘
                   │
┌──────────────────▼──────────────────────────────────┐
│              Data Processing Layer                   │
│  Python Scripts: 清洗 / 异常检测 / 趋势分析           │
└──────────────────┬──────────────────────────────────┘
                   │
┌──────────────────▼──────────────────────────────────┐
│             Data Acquisition Layer                   │
│         Pangolinfo Scrape API (99%+ 成功率)           │
│    Amazon US/UK/DE/JP + Walmart + eBay + Etsy        │
└─────────────────────────────────────────────────────┘

环境依赖

pip install requests langchain openai pandas schedule python-dotenv

# .env 配置
PANGOLINFO_API_KEY=your_pangolinfo_key
OPENAI_API_KEY=your_openai_key
SLACK_WEBHOOK_URL=https://hooks.slack.com/services/xxx/yyy/zzz

核心模块实现

1. 数据采集模块

# amazon_data_client.py
import requests
import os
from typing import Optional

class AmazonDataClient:
    """Pangolinfo Scrape API 客户端封装"""
    
    BASE_URL = "https://api.pangolinfo.com/v1/amazon"
    
    def __init__(self, api_key: Optional[str] = None):
        self.api_key = api_key or os.getenv("PANGOLINFO_API_KEY")
        self.session = requests.Session()
        self.session.headers.update({"Authorization": f"Bearer {self.api_key}"})
    
    def get_product(self, asin: str, marketplace: str = "US") -> dict:
        """获取商品实时数据（BSR、价格、库存、评论）"""
        resp = self.session.get(
            f"{self.BASE_URL}/product",
            params={
                "asin": asin,
                "marketplace": marketplace,
                "fields": "bsr,price,inventory_status,review_count,title,brand"
            },
            timeout=15
        )
        resp.raise_for_status()
        return resp.json()
    
    def get_category_top(self, category_id: str, limit: int = 100) -> list:
        """获取品类Top N ASIN列表"""
        resp = self.session.get(
            f"{self.BASE_URL}/category/bestsellers",
            params={"category_id": category_id, "marketplace": "US", "limit": limit},
            timeout=30
        )
        resp.raise_for_status()
        return resp.json().get("asins", [])
    
    def get_search_results(self, keyword: str, pages: int = 3) -> list:
        """获取关键词搜索结果页数据"""
        results = []
        for page in range(1, pages + 1):
            resp = self.session.get(
                f"{self.BASE_URL}/search",
                params={"keyword": keyword, "marketplace": "US", "page": page},
                timeout=20
            )
            if resp.ok:
                results.extend(resp.json().get("results", []))
        return results

2. 异常检测模块

# anomaly_detector.py
from dataclasses import dataclass
from typing import List

@dataclass
class Alert:
    alert_type: str
    asin: str
    message: str
    severity: str  # HIGH / MEDIUM / LOW
    action_suggestion: str

class BSRAnomalyDetector:
    """BSR异常检测：断货预警 + 需求激增检测"""
    
    LOW_STOCK_SIGNALS = [
        "Only 1 left in stock",
        "Only 2 left in stock", 
        "Only 3 left in stock",
        "Temporarily out of stock",
        "Currently unavailable"
    ]
    
    def __init__(self, bsr_surge_threshold: int = 200, window_size: int = 4):
        self.bsr_surge_threshold = bsr_surge_threshold
        self.window_size = window_size  # 检测窗口（采样次数）
        self.bsr_history: dict = {}
    
    def process(self, asin: str, current_bsr: int, inv_status: str) -> List[Alert]:
        alerts = []
        
        # 断货预警
        if any(sig in inv_status for sig in self.LOW_STOCK_SIGNALS):
            alerts.append(Alert(
                alert_type="STOCKOUT_IMMINENT",
                asin=asin,
                message=f"库存预警: {inv_status}",
                severity="HIGH",
                action_suggestion="立即将该品类广告预算提升40-60%，窗口期预计12-36小时"
            ))
        
        # BSR激增检测（排名数字下降 = 销量上升）
        history = self.bsr_history.get(asin, [])
        if len(history) >= 2:
            prev_bsr = history[-1]
            bsr_change = prev_bsr - current_bsr  # 正值=排名上升
            if bsr_change > self.bsr_surge_threshold:
                alerts.append(Alert(
                    alert_type="BSR_SURGE",
                    asin=asin,
                    message=f"BSR单次上升{bsr_change}位，当前排名#{current_bsr}",
                    severity="MEDIUM",
                    action_suggestion="分析需求驱动因素，评估是否需要紧急补货"
                ))
        
        # 更新历史
        history.append(current_bsr)
        self.bsr_history[asin] = history[-self.window_size:]
        
        return alerts

3. AI Agent工具集

# agent_tools.py
from langchain.tools import tool
from amazon_data_client import AmazonDataClient
import json

client = AmazonDataClient()

@tool
def get_asin_realtime_data(asin: str) -> str:
    """
    获取指定ASIN的亚马逊实时数据，包括BSR排名、价格、库存状态和评论数量。
    输入: ASIN字符串（例如 B0XXXXXXXXX）
    输出: JSON格式的实时数据
    """
    data = client.get_product(asin)
    return json.dumps(data, ensure_ascii=False)

@tool
def scan_patent_risk(keyword: str) -> str:
    """
    扫描指定关键词在USPTO/WIPO数据库中的专利风险。
    输入: 产品描述关键词（英文）
    输出: 风险等级（LOW/MEDIUM/HIGH）和相关专利摘要
    """
    # 示意实现——实际接入USPTO公开API或专利扫描服务
    return json.dumps({
        "keyword": keyword,
        "risk_level": "MEDIUM",
        "note": "发现2项相关外观设计专利，建议审查设计差异化方案",
        "patents_found": 2
    }, ensure_ascii=False)

@tool
def get_category_opportunities(category_keyword: str) -> str:
    """
    分析指定品类关键词的市场机会，返回Top20 ASIN的BSR分布和竞争格局。
    输入: 品类关键词（英文）
    输出: 市场分析报告JSON
    """
    results = client.get_search_results(category_keyword, pages=2)
    analysis = {
        "keyword": category_keyword,
        "total_asins_analyzed": len(results),
        "top_bsr_range": "需要实际数据",
        "competition_level": "MEDIUM",
        "entry_recommendation": "中等竞争，建议差异化切入"
    }
    return json.dumps(analysis, ensure_ascii=False)

4. LangChain AI Agent主程序

# main_agent.py
from langchain.agents import initialize_agent, AgentType
from langchain.chat_models import ChatOpenAI
from agent_tools import get_asin_realtime_data, scan_patent_risk, get_category_opportunities
import os

def create_product_research_agent():
    """创建亚马逊选品研究AI Agent"""
    llm = ChatOpenAI(
        model="gpt-4o",
        temperature=0,
        openai_api_key=os.getenv("OPENAI_API_KEY")
    )
    
    tools = [
        get_asin_realtime_data,
        scan_patent_risk,
        get_category_opportunities
    ]
    
    agent = initialize_agent(
        tools=tools,
        llm=llm,
        agent=AgentType.OPENAI_FUNCTIONS,
        verbose=True,
        max_iterations=10
    )
    return agent

if __name__ == "__main__":
    agent = create_product_research_agent()
    
    # 示例任务：分析POD装饰画品类机会
    task = """
    分析以下任务：
    1. 搜索关键词 "3D sandstone relief wall art" 的品类竞争情况
    2. 扫描该品类的专利风险
    3. 如果找到候选ASIN，获取其实时BSR数据
    4. 综合以上数据，给出入场建议（风险等级、推荐入场价区间、备货时机）
    """
    
    result = agent.run(task)
    print("\n=== 选品分析报告 ===")
    print(result)

5. 定时监控调度

# scheduler.py
import schedule
import time
import json
from datetime import datetime
from amazon_data_client import AmazonDataClient
from anomaly_detector import BSRAnomalyDetector
import requests
import os

client = AmazonDataClient()
detector = BSRAnomalyDetector(bsr_surge_threshold=200)

WATCH_LIST = [
    {"asin": "B0XXXXXXXXX", "label": "竞品A"},
    {"asin": "B0YYYYYYYYY", "label": "竞品B"},
]

def send_slack_alert(alerts):
    """发送Slack预警通知"""
    webhook_url = os.getenv("SLACK_WEBHOOK_URL")
    if not webhook_url or not alerts:
        return
    
    message = "\n".join([
        f"🚨 [{a.severity}] {a.alert_type} | ASIN: {a.asin}\n"
        f"📊 {a.message}\n💡 {a.action_suggestion}"
        for a in alerts
    ])
    
    requests.post(webhook_url, json={"text": f"亚马逊实时监控预警\n{message}"})

def monitoring_cycle():
    """单次监控循环"""
    print(f"[{datetime.now().strftime('%H:%M:%S')}] 开始监控...")
    all_alerts = []
    
    for item in WATCH_LIST:
        try:
            data = client.get_product(item["asin"])
            alerts = detector.process(
                asin=item["asin"],
                current_bsr=data.get("bsr", 0),
                inv_status=data.get("inventory_status", "")
            )
            all_alerts.extend(alerts)
        except Exception as e:
            print(f"Error fetching {item['asin']}: {e}")
    
    if all_alerts:
        send_slack_alert(all_alerts)
        print(f"发送 {len(all_alerts)} 条预警")

# 每30分钟执行一次监控
schedule.every(30).minutes.do(monitoring_cycle)

if __name__ == "__main__":
    print("亚马逊实时数据监控系统已启动（30分钟间隔）")
    monitoring_cycle()  # 立即执行第一次
    while True:
        schedule.run_pending()
        time.sleep(60)

部署建议

# 推荐用 supervisor 守护进程
[program:amazon_monitor]
command=python /path/to/scheduler.py
directory=/path/to/project
autostart=true
autorestart=true
stderr_logfile=/var/log/amazon_monitor.err.log
stdout_logfile=/var/log/amazon_monitor.out.log

核心参数调优参考

参数	保守配置	激进配置	说明
BSR采样间隔	60分钟	15分钟	越短API消耗越大
断货预警阈值	“Only 3 left”	“Only 5 left”	越早预警越多误报
BSR激增阈值	200位	100位	根据品类BSR分布调整
WATCH_LIST大小	20-50 ASIN	100-200 ASIN	影响月API成本

---

如果这套架构对你有帮助，欢迎点赞收藏。有技术问题可以在评论区交流，或者去Pangolinfo文档中心提工单。