使用LangGraph构建自主RAG(3)

===

设置WebSearch API:

我们使用serperAPI进行网络搜索(谷歌搜索)。其免费层提供2500次免费网络搜索API。它能在1 - 2秒内为给定查询提供谷歌搜索结果。

你可以在这里找到更多详细信息并获取你的 API 密钥此处

ini 复制代码
from langchain_community.utilities import GoogleSerperAPIWrapper
from dotenv import load_dotenv
load_dotenv()
SERPER_API_KEY = os.getenv("SERPER_API_KEY")

search = GoogleSerperAPIWrapper()
# testing the google search API
search.run(query="What are the various vaccines of COVID-19")

输出:

arduino 复制代码
"Different types of COVID-19 vaccines: How they work · mRNA vaccine · mRNA vaccine · Messenger RNA (mRNA) vaccine · Viral vector vaccine · Viral vector vaccine ... First introduced in December 2020, the original COVID mRNA vaccines from both Pfizer and Moderna protected against the original SARS-CoV-2 virus ... Currently, there are two types of COVID-19 vaccines for use in the United States: mRNA, and protein subunit vaccines. None of these vaccines can ... On this page, you will find infographics to explain how different types of vaccines work, including the Pfizer/BioNTech vaccine, the Moderna vaccine and the ... Types of COVID-19 vaccines. Two types of COVID-19 vaccines are recommended for use in the United States: mRNA vaccines. Moderna COVID-19 Vaccine ... List of COVID-19 vaccine authorizations · Contents · Overview maps · Oxford--AstraZeneca · Pfizer--BioNTech · Janssen · Moderna · Sinopharm BIBP · Sputnik V. WHO recommends a simplified single-dose regime for primary immunization for most COVID-19 vaccines which would improve acceptance and uptake and provide ... Find information and answers to your questions about the COVID-19 vaccine, including scheduling, kid's shots, boosters, additional doses, records and more. The California Department of Public Health is dedicated to optimizing the health and well-being of Californians. Find out how the COVID-19 vaccines work, how many doses are needed, possible side effects and who shouldn't get the vaccine."

设置OpenAI API客户端:

我们将使用Open AI API的gpt-5-nano模型进行本次实验。

我选择这个模型是因为我的实验规模很小且复杂度较低。我的提示非常简单。在现实场景中,你可能需要使用更高级的模型(gpt-5-mini / gpt-5)。

参考以下代码:

python 复制代码
from openai import OpenAI
from dotenv import load_dotenv
load_dotenv()
openai_api_key = os.getenv("OPEN_AI_KEY")
def get_llm_response(prompt: str) -> str:
    """Function to get response from LLM"""
    from openai import OpenAI
    client_llm = OpenAI(api_key=openai_api_key)
    response = client_llm.chat.completions.create(
        model="gpt-5-nano",
        messages=[{"role": "user", "content": prompt}]
    )
    return response.choices[0].message.content

## testing the LLM response
prompt = "Explain the theory of relativity in simple terms in 30 words"
response = get_llm_response(prompt)
print(response)

相对论认为,物理定律适用于所有观察者,不存在优先参考系,且没有任何物体的运动速度能超过光速。时间和空间会随运动和引力而弯曲,从根本上使质量和能量等价。

现在场地已经布置好了。让我们开始实施RAG。首先,我们将使用第一个来源(医疗问答)构建一个传统RAG,然后再转向智能体RAG。

相关推荐
心枢AI研习社10 分钟前
我问了claude目前最强大的模型fable 5这个问题?
人工智能·agent·claude
明略科技12 分钟前
什么是 RAG?为什么光靠大模型的记忆力远远不够
开源·agent
七牛开发者35 分钟前
Skills 是什么?Claude 官方教你做一个好用的 Skill
aigc·openai·claude
七牛开发者1 小时前
AI Agent 的 4 个工程关键词:Prompt、Context、Loop、Harness 到底是什么?
aigc·openai·agent
付玉祥1 小时前
Agent 开发:你真的需要框架吗?
agent
小七-七牛开发者2 小时前
AI Agent 的 4 个工程关键词:Prompt、Context、Loop、Harness 到底是什么?
ai·大模型·agent·token·context·loop·codex·harness
用户5191495848453 小时前
CVE-2025-0282 Ivanti 远程命令执行漏洞利用工具
人工智能·aigc
是发财不是旺财3 小时前
Hermes 网关四层权限控制方案:让 AI Agent 安全地查数据库
数据库·安全·agent·openclaw·hermes
猫头_4 小时前
跨 AI IDE 的协作痛点:用了五六个 AI 编辑器,提示词(Skills)该怎么统一管理?
openai·ai编程·cursor
92year4 小时前
MCP STDIO 命令注入:一个架构决策引发的 RCE 连锁反应
aigc