Python 构建 AI 助手：文档总结、代码生成与智能检索实战

1.3 核心工具类封装

在开始之前，我们先封装一个统一的 LLM 调用类。这一步很关键，它能避免后续业务逻辑被 API 细节耦合。

import os
import asyncio
from typing import List, Dict, Optional, AsyncGenerator
from dataclasses import dataclass
from openai import AsyncOpenAI
from dotenv import load_dotenv

load_dotenv()

@dataclass
class Message:
    """消息数据结构"""
    role: str  # system / user / assistant
    content: str

class LLMClient:
    """统一的大模型客户端"""
    def __init__(
        self,
        api_key: str = None,
        base_url: str = None,
        model: str = "deepseek-chat",
        temperature: float = 0.7
    ):
        self.api_key = api_key or os.getenv("DEEPSEEK_API_KEY")
        self.base_url = base_url or os.getenv("DEEPSEEK_BASE_URL")
        self.model = model
        self.temperature = temperature
        self.client = AsyncOpenAI(
            api_key=self.api_key,
            base_url=self.base_url
        )

    async def chat(
        self,
        messages: List[Message],
        stream: bool = False,
        **kwargs
    ) -> str:
        """发送聊天请求
        Args:
            messages: 消息列表
            stream: 是否流式输出
            **kwargs: 其他参数（max_tokens 等）
        Returns:
            模型回复内容
        """
        response = await self.client.chat.completions.create(
            model=self.model,
            messages=[{"role": m.role, "content": m.content} for m in messages],
            temperature=kwargs.get("temperature", self.temperature),
            stream=stream,
            max_tokens=kwargs.get("max_tokens", 4000)
        )
        if stream:
            full_content = ""
            async for chunk in response:
                if chunk.choices[0].delta.content:
                    content = chunk.choices[0].delta.content
                    full_content += content
                    print(content, end="", flush=True)
            return full_content
        else:
            return response.choices[0].message.content

    async def chat_with_functions(
        self,
        messages: List[Message],
        functions: List[Dict]
    ) -> Dict:
        """带函数调用的聊天（用于代码执行等场景）"""
        response = await self.client.chat.completions.create(
            model=self.model,
            messages=[{"role": m.role, "content": m.content} for m in messages],
            tools=functions,
            tool_choice="auto"
        )
        return response.choices[0].message

# 使用示例
async def test_llm():
    llm = LLMClient()
    response = await llm.chat([
        Message(role="user", content="用 Python 写一个快速排序")
    ])
    print(response)

if __name__ == "__main__":
    asyncio.run(test_llm())

二、工具一：智能文档总结器

2.1 功能设计

这个工具的核心在于处理长文本。直接丢给大模型会超出 Token 限制，所以我们需要分块处理，再合并摘要。

• 输入：PDF、Word、网页、Markdown
• 解析：根据类型调用不同解析库
• 分块：按 Token 数或段落切分，保留重叠上下文
• 并行总结：每个块独立处理，提升速度
• 二次总结：合并块摘要，形成最终结果

2.2 核心代码实现

这里重点看 TextChunker 和 DocumentSummarizer。注意我们使用了 asyncio.Semaphore 来控制并发数，防止瞬间请求过多导致限流。

import asyncio
from typing import List, Optional
from pathlib import Path
import PyPDF2
from bs4 import BeautifulSoup
import aiohttp
from dataclasses import dataclass
from datetime import datetime

@dataclass
class DocumentSummary:
    """文档摘要结果"""
    title: str
    summary: str
    key_points: List[str]
    reading_time: int  # 预计阅读时间（分钟）
    word_count: int
    created_at: str

class DocumentParser:
    """文档解析器"""
    @staticmethod
    async def parse_pdf(file_path: str) -> str:
        """解析 PDF 文件"""
        text = ""
        with open(file_path, 'rb') as file:
            pdf_reader = PyPDF2.PdfReader(file)
            for page in pdf_reader.pages:
                text += page.extract_text() + "\n"
        return text

    @staticmethod
    async def parse_text(file_path: str) -> str:
        """解析纯文本文件"""
        with open(file_path, 'r', encoding='utf-8') as f:
            return f.read()

    @staticmethod
    async def parse_url(url: str) -> str:
        """解析网页内容"""
        async with aiohttp.ClientSession() as session:
            async with session.get(url) as response:
                html = await response.text()
                soup = BeautifulSoup(html, 'html.parser')
                # 移除脚本和样式
                for script in soup(['script', 'style']):
                    script.decompose()
                return soup.get_text(separator='\n', strip=True)

class TextChunker:
    """文本分块器"""
    def __init__(self, chunk_size: int = 3000, overlap: int = 200):
        self.chunk_size = chunk_size
        self.overlap = overlap

    def chunk(self, text: str) -> List[str]:
        """将文本分成多个块"""
        paragraphs = text.split('\n\n')
        chunks = []
        current_chunk = ""
        for para in paragraphs:
            if len(current_chunk) + len(para) <= self.chunk_size:
                current_chunk += para + "\n\n"
            else:
                if current_chunk:
                    chunks.append(current_chunk.strip())
                if len(para) > self.chunk_size:
                    for i in range(0, len(para), self.chunk_size - self.overlap):
                        chunks.append(para[i:i + self.chunk_size])
                current_chunk = ""
        if current_chunk:
            chunks.append(current_chunk.strip())
        return chunks

class DocumentSummarizer:
    """智能文档总结器"""
    def __init__(self, llm_client: LLMClient):
        self.llm = llm_client
        self.parser = DocumentParser()
        self.chunker = TextChunker()

    async def summarize(
        self,
        source: str,
        source_type: str = "file",
        output_format: str = "markdown"
    ) -> DocumentSummary:
        """总结文档"""
        print(f"📖 正在解析文档：{source}")
        
        # 1. 解析文档
        if source_type == "url":
            text = await self.parser.parse_url(source)
            title = await self._extract_title_from_url(text)
        else:
            if source.endswith('.pdf'):
                text = await self.parser.parse_pdf(source)
            else:
                text = await self.parser.parse_text(source)
            title = Path(source).stem
            
        word_count = len(text)
        reading_time = max(1, word_count // 500)
        print(f"✅ 解析完成，共 {word_count} 字，预计阅读 {reading_time} 分钟")
        
        # 2. 分块
        chunks = self.chunker.chunk(text)
        print(f"🔪 正在分块... 分成 {len(chunks)} 个块")
        
        # 3. 并行总结每个块
        print(f"🤖 正在 AI 总结...")
        chunk_summaries = await self._summarize_chunks(chunks)
        
        # 4. 二次总结
        print(f"🔄 正在整合摘要...")
        final_summary = await self._merge_summaries(chunk_summaries, title)
        
        # 5. 提取关键要点
        key_points = await self._extract_key_points(final_summary)
        
        return DocumentSummary(
            title=title,
            summary=final_summary,
            key_points=key_points,
            reading_time=reading_time,
            word_count=word_count,
            created_at=datetime.now().strftime("%Y-%m-%d %H:%M:%S")
        )

    async def _summarize_chunks(self, chunks: List[str]) -> List[str]:
        """并行总结每个文本块"""
        semaphore = asyncio.Semaphore(5)  # 限制并发数
        
        async def summarize_chunk(chunk: str, index: int):
            async with semaphore:
                prompt = f"""请总结以下文本的核心内容，要求：
                1. 保留关键信息（数据、结论、人名等）
                2. 省略细节和例子
                3. 用简洁的语言表达
                4. 200 字以内
                文本内容：{chunk}
                总结："""
                response = await self.llm.chat([
                    Message(role="system", content="你是一个专业的内容总结助手"),
                    Message(role="user", content=prompt)
                ])
                print(f" └─ 块 {index+1}/{len(chunks)} 完成")
                return response

        tasks = [summarize_chunk(chunk, i) for i, chunk in enumerate(chunks)]
        return await asyncio.gather(*tasks)

    async def _merge_summaries(self, summaries: List[str], title: str) -> str:
        """合并所有摘要"""
        combined = "\n\n".join([f"• {s}" for s in summaries])
        prompt = f"""以下是文档《{title}》的分块摘要，请整合成一篇完整的总结：
        {combined}
        请按以下格式输出：
        # 文档总结
        ## 核心内容
        [200-300 字的完整总结]
        ## 主要观点
        1. [观点 1]
        2. [观点 2]
        ...
        整合后的总结："""
        response = await self.llm.chat([
            Message(role="system", content="你是一个专业的内容整合助手"),
            Message(role="user", content=prompt)
        ])
        return response

    async def _extract_key_points(self, summary: str) -> List[str]:
        """提取关键要点"""
        prompt = f"""从以下总结中提取 5-7 个关键要点，每点不超过 20 字：
        {summary}
        只输出要点列表，每行一个："""
        response = await self.llm.chat([
            Message(role="user", content=prompt)
        ])
        return [line.strip() for line in response.split('\n') if line.strip()]

    async def _extract_title_from_url(self, text: str) -> str:
        """从网页文本中提取标题"""
        prompt = f"""从以下文本中提取文章标题，只返回标题：
        {text[:500]}
        标题："""
        response = await self.llm.chat([
            Message(role="user", content=prompt)
        ])
        return response.strip()

# 使用示例
async def main_summarizer():
    llm = LLMClient()
    summarizer = DocumentSummarizer(llm)
    result = await summarizer.summarize(
        source="research_paper.pdf",
        source_type="file"
    )
    print("\n" + "="*60)
    print(f"📄 标题：{result.title}")
    print(f"⏱️ 预计阅读时间：{result.reading_time} 分钟")
    print(f"📊 字数：{result.word_count}")
    print("\n🔑 关键要点:")
    for point in result.key_points:
        print(f" • {point}")
    print(f"\n📝 总结:\n{result.summary}")

if __name__ == "__main__":
    asyncio.run(main_summarizer())

2.3 使用效果对比

三、工具二：AI 代码生成器

3.1 功能架构

除了生成新代码，这个工具还能解释现有代码、优化性能甚至修复 Bug。关键在于意图识别和安全检查。

• 生成模式：根据自然语言描述生成可运行代码
• 解释模式：分析代码逻辑和复杂度
• 优化模式：提升性能和可读性
• 调试模式：结合错误日志定位问题
• 测试模式：自动生成 pytest 用例

3.2 核心实现

这里有一个重要的细节：我们在生成代码后增加了安全检查，比如检测是否包含 eval() 或 exec()，这在自动化场景中非常必要。

import re
import subprocess
import tempfile
from typing import Dict, List, Optional, Tuple
from enum import Enum
import ast

class CodeMode(Enum):
    """代码生成模式"""
    GENERATE = "generate"
    EXPLAIN = "explain"
    OPTIMIZE = "optimize"
    DEBUG = "debug"
    TEST = "test"

@dataclass
class CodeResult:
    """代码生成结果"""
    code: str
    language: str
    explanation: str
    tests: Optional[str] = None
    warnings: List[str] = None

class CodeGenerator:
    """AI 代码生成器"""
    def __init__(self, llm_client: LLMClient):
        self.llm = llm_client
        # 代码质量检查规则
        self.quality_rules = {
            "security": [
                r"eval\s*\(",
                r"exec\s*\(",
                r"pickle\.loads?"
            ],
            "performance": [
                r"for\s+\w+\s+in\s+range\(len\("
            ]
        }

    async def generate(
        self,
        requirement: str,
        language: str = "python",
        mode: CodeMode = CodeMode.GENERATE,
        context: str = ""
    ) -> CodeResult:
        """生成/处理代码"""
        mode_prompts = {
            CodeMode.GENERATE: self._build_generate_prompt,
            CodeMode.EXPLAIN: self._build_explain_prompt,
            CodeMode.OPTIMIZE: self._build_optimize_prompt,
            CodeMode.DEBUG: self._build_debug_prompt,
            CodeMode.TEST: self._build_test_prompt,
        }
        prompt_builder = mode_prompts[mode]
        prompt = prompt_builder(requirement, language, context)
        print(f"🤖 正在生成{mode.value}...")
        
        response = await self.llm.chat([
            Message(role="system", content=self._get_system_prompt(language)),
            Message(role="user", content=prompt)
        ])
        
        code, explanation = self._parse_code_response(response, language)
        warnings = self._security_check(code)
        
        tests = None
        if mode == CodeMode.GENERATE:
            tests = await self._generate_tests(code, language)
            
        return CodeResult(
            code=code,
            language=language,
            explanation=explanation,
            tests=tests,
            warnings=warnings
        )

    def _get_system_prompt(self, language: str) -> str:
        return f"""你是一个专业的{language}程序员和教师。
        输出代码时：
        1. 代码必须可直接运行
        2. 添加必要的注释和文档字符串
        3. 遵循{language}最佳实践和 PEP8 规范
        4. 包含错误处理
        5. 代码后附上简洁的使用说明
        输出格式：
        ```python
        # 代码块
        ```
        **使用说明：**
        [说明内容]
        """

    def _build_generate_prompt(
        self,
        requirement: str,
        language: str,
        context: str
    ) -> str:
        if context:
            return f"""请根据以下需求生成{language}代码：
            需求：{requirement}
            上下文代码：
            ```
            {context}
            ```
            请生成完整的、可直接运行的代码。"""
        return f"""请根据以下需求生成{language}代码：
        需求：{requirement}
        要求：
        1. 代码完整且可运行
        2. 包含必要的输入验证和错误处理
        3. 添加清晰的注释
        4. 如果是算法，注明时间复杂度
        请生成代码："""

    def _build_explain_prompt(
        self,
        code: str,
        language: str,
        context: str
    ) -> str:
        return f"""请详细解释以下{language}代码的功能和工作原理：
        ```
        {code}
        ```
        请从以下几个方面解释：
        1. 整体功能概述
        2. 关键代码逻辑
        3. 使用的数据结构和算法
        4. 时间/空间复杂度
        5. 可能的改进点
        详细解释："""

    def _build_optimize_prompt(
        self,
        code: str,
        language: str,
        context: str
    ) -> str:
        return f"""请优化以下{language}代码：
        ```
        {code}
        ```
        优化目标：
        1. 提升性能
        2. 改善可读性
        3. 增强健壮性
        4. 遵循最佳实践
        请给出：
        1. 优化后的代码
        2. 优化点说明
        优化结果："""

    def _build_debug_prompt(
        self,
        code: str,
        language: str,
        context: str
    ) -> str:
        return f"""请分析以下{language}代码中的问题并修复：
        ```
        {code}
        ```
        可能的错误信息：
        {context if context else "[无]"}
        请给出：
        1. 问题分析
        2. 修复后的代码
        3. 预防建议
        分析结果："""

    def _build_test_prompt(
        self,
        code: str,
        language: str,
        context: str
    ) -> str:
        return f"""请为以下{language}代码生成完整的测试用例：
        ```
        {code}
        ```
        测试要求：
        - 测试函数名以 test_开头
        - 包含正常和异常情况
        - 使用合适的测试框架（如 pytest）
        只输出测试代码："""

    def _parse_code_response(self, response: str, language: str) -> Tuple[str, str]:
        """解析 LLM 响应，提取代码和说明"""
        code_pattern = rf"```{language}\n(.*?)```"
        code_match = re.search(code_pattern, response, re.DOTALL)
        if code_match:
            code = code_match.group(1).strip()
            explanation = response.replace(code_match.group(0), "").strip()
        else:
            code = response
            explanation = "无额外说明"
        return code, explanation

    def _security_check(self, code: str) -> List[str]:
        """代码安全检查"""
        warnings = []
        for category, patterns in self.quality_rules.items():
            for pattern in patterns:
                if re.search(pattern, code):
                    warnings.append(f"⚠️ 安全警告：检测到 {pattern} 使用")
        try:
            ast.parse(code)
        except SyntaxError as e:
            warnings.append(f"⚠️ 语法错误：{e}")
        return warnings

    async def _generate_tests(self, code: str, language: str) -> str:
        """生成测试代码"""
        prompt = f"""为以下{language}代码编写 pytest 测试：
        ```
        {code}
        ```
        要求：
        - 测试函数名以 test_开头
        - 包含正常和异常情况
        - 使用 pytest 断言
        只输出测试代码："""
        response = await self.llm.chat([
            Message(role="user", content=prompt)
        ])
        return response

    async def execute_code(
        self,
        code: str,
        language: str = "python",
        timeout: int = 10
    ) -> Dict:
        """安全执行代码"""
        with tempfile.NamedTemporaryFile(
            mode='w', suffix=f'.{language}', delete=False
        ) as f:
            f.write(code)
            temp_file = f.name
        try:
            result = subprocess.run(
                ['python', temp_file],
                capture_output=True,
                text=True,
                timeout=timeout
            )
            return {
                "success": result.returncode == 0,
                "output": result.stdout,
                "error": result.stderr
            }
        except subprocess.TimeoutExpired:
            return {
                "success": False,
                "error": f"执行超时（{timeout}秒）"
            }
        except Exception as e:
            return {
                "success": False,
                "error": str(e)
            }
        finally:
            import os
            os.unlink(temp_file)

# 交互式代码助手
class InteractiveCodeAssistant:
    """交互式代码助手"""
    def __init__(self, llm_client: LLMClient):
        self.generator = CodeGenerator(llm_client)
        self.history: List[Dict] = []

    async def chat(self, user_input: str) -> str:
        """对话式代码助手"""
        intent = await self._detect_intent(user_input)
        if intent == "generate":
            result = await self.generator.generate(
                requirement=user_input, mode=CodeMode.GENERATE
            )
            output = f"```python\n{result.code}\n```\n\n"
            output += f"**说明：**\n{result.explanation}\n\n"
            if result.warnings:
                output += "**安全警告：**\n" + "\n".join(result.warnings) + "\n\n"
            if result.tests:
                output += f"**测试代码：**\n```python\n{result.tests}\n```"
            return output
        elif intent == "explain":
            code = self._extract_code_from_input(user_input)
            result = await self.generator.generate(
                requirement=code, mode=CodeMode.EXPLAIN
            )
            return result.explanation

    async def _detect_intent(self, user_input: str) -> str:
        """检测用户意图"""
        prompt = f"""判断用户意图，只返回：generate / explain / optimize / debug
        用户输入：{user_input}
        意图："""
        response = await self.generator.llm.chat([
            Message(role="user", content=prompt)
        ])
        intent = response.strip().lower()
        return intent if intent in ["generate", "explain", "optimize", "debug"] else "generate"

    def _extract_code_from_input(self, user_input: str) -> str:
        """从输入中提取代码"""
        match = re.search(r'```(?:python)?\n(.*?)```', user_input, re.DOTALL)
        if match:
            return match.group(1).strip()
        return user_input

# 使用示例
async def main_code_generator():
    llm = LLMClient()
    assistant = InteractiveCodeAssistant(llm)

    # 示例 1：生成代码
    print("="*60)
    print("示例 1：生成快速排序代码")
    print("="*60)
    result = await assistant.chat("用 Python 实现一个快速排序，要求有详细注释")
    print(result)

    # 示例 2：解释代码
    print("\n" + "="*60)
    print("示例 2：解释代码")
    print("="*60)
    code = """
def quicksort(arr):
    if len(arr) <= 1:
        return arr
    pivot = arr[len(arr) // 2]
    left = [x for x in arr if x < pivot]
    middle = [x for x in arr if x == pivot]
    right = [x for x in arr if x > pivot]
    return quicksort(left) + middle + quicksort(right)
"""
    explanation = await assistant.chat(f"解释这段代码在做什么：\n`{code}`")
    print(explanation)

if __name__ == "__main__":
    asyncio.run(main_code_generator())

3.3 代码生成能力对比

四、工具三：智能资料助手

4.1 系统架构

这个工具不仅仅是搜索，它还会聚合多源信息（搜索引擎、官方文档、StackOverflow），然后由 AI 进行综合推理。

• 问题分析：判断是事实查询还是技术文档查询
• 多源并行搜索：同时调用 Bing、DuckDuckGo、SO
• 内容清洗：去除广告和无关 HTML 标签
• 相关性排序：基于来源权重排序
• AI 总结整合：生成结构化答案并标注引用

4.2 核心代码

注意 SearchEngine 类中实现了备用方案，如果没有配置 Bing Key，会自动降级到 DuckDuckGo，保证工具可用性。

import aiohttp
from typing import List, Dict, Optional
from dataclasses import dataclass
import re
from urllib.parse import quote, urljoin
import json
import os

@dataclass
class SearchResult:
    """搜索结果"""
    title: str
    url: str
    snippet: str
    source: str  # google / bing / docs / stackoverflow
    relevance: float = 0.0

@dataclass
class ResearchResult:
    """研究结果"""
    answer: str
    sources: List[SearchResult]
    related_questions: List[str]
    confidence: float

class SearchEngine:
    """搜索引擎封装"""
    def __init__(self, bing_api_key: str = None):
        self.bing_api_key = bing_api_key or os.getenv("BING_SEARCH_API_KEY")
        self.headers = {"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36"}

    async def search_bing(self, query: str, count: int = 10) -> List[SearchResult]:
        """使用 Bing 搜索 API"""
        if not self.bing_api_key:
            return await self._search_duckduckgo(query, count)
        url = "https://api.bing.microsoft.com/v7.0/search"
        params = {"q": query, "count": count, "responseFilter": "webpages"}
        async with aiohttp.ClientSession() as session:
            async with session.get(
                url, params=params, headers={"Ocp-Apim-Subscription-Key": self.bing_api_key}
            ) as response:
                data = await response.json()
                results = []
                for item in data.get("webPages", {}).get("value", []):
                    results.append(SearchResult(
                        title=item["name"],
                        url=item["url"],
                        snippet=item["snippet"],
                        source="bing"
                    ))
                return results

    async def _search_duckduckgo(self, query: str, count: int = 10) -> List[SearchResult]:
        """使用免费的 DuckDuckGo 搜索"""
        url = f"https://html.duckduckgo.com/html/?q={quote(query)}"
        async with aiohttp.ClientSession() as session:
            async with session.get(url, headers=self.headers) as response:
                html = await response.text()
                from bs4 import BeautifulSoup
                soup = BeautifulSoup(html, 'html.parser')
                results = []
                for result in soup.select('.result')[:count]:
                    title_elem = result.select_one('.result__a')
                    snippet_elem = result.select_one('.result__snippet')
                    url_elem = result.select_one('.result__url')
                    if title_elem and url_elem:
                        results.append(SearchResult(
                            title=title_elem.get_text(),
                            url=url_elem.get('href', ''),
                            snippet=snippet_elem.get_text() if snippet_elem else '',
                            source="duckduckgo"
                        ))
                return results

    async def search_stackoverflow(self, query: str, count: int = 5) -> List[SearchResult]:
        """搜索 StackOverflow"""
        search_query = f"site:stackoverflow.com {query}"
        results = await self._search_duckduckgo(search_query, count)
        for r in results:
            r.source = "stackoverflow"
        return results

    async def search_docs(self, query: str, docs_domain: str, count: int = 5) -> List[SearchResult]:
        """搜索特定文档站（如 Python 文档）"""
        search_query = f"site:{docs_domain}{query}"
        results = await self._search_duckduckgo(search_query, count)
        for r in results:
            r.source = "docs"
        return results

class IntelligentResearcher:
    """智能研究助手"""
    def __init__(self, llm_client: LLMClient, search_engine: SearchEngine):
        self.llm = llm_client
        self.search = search_engine

    async def research(
        self,
        question: str,
        depth: int = 1,
        sources: List[str] = None
    ) -> ResearchResult:
        """研究问题"""
        print(f"🔍 正在研究：{question}")
        
        # 1. 并行搜索多个源
        search_tasks = []
        if not sources or "google" in sources:
            search_tasks.append(self.search.search_bing(question))
        if not sources or "stackoverflow" in sources:
            search_tasks.append(self.search.search_stackoverflow(question))
        
        # 如果是技术问题，搜索官方文档
        if self._is_technical_question(question):
            tech = await self._detect_tech_stack(question)
            if tech:
                docs_url = self._get_docs_url(tech)
                search_tasks.append(self.search.search_docs(question, docs_url))
        
        search_results_list = await asyncio.gather(*search_tasks)
        all_results = []
        for results in search_results_list:
            all_results.extend(results)
        print(f"📊 找到 {len(all_results)} 条相关结果")
        
        # 2. 提取页面内容（深度研究）
        if depth > 1:
            all_results = await self._fetch_page_contents(all_results[:5])
        
        # 3. AI 分析并整合答案
        answer = await self._synthesize_answer(question, all_results)
        
        # 4. 生成相关问题
        related = await self._generate_related_questions(question, answer)
        
        # 5. 计算置信度
        confidence = self._calculate_confidence(all_results)
        
        return ResearchResult(
            answer=answer,
            sources=all_results[:5],
            related_questions=related,
            confidence=confidence
        )

    def _is_technical_question(self, question: str) -> bool:
        """判断是否是技术问题"""
        tech_keywords = ["python", "javascript", "java", "api", "函数", "如何使用", "怎么用", "documentation", "example"]
        return any(kw in question.lower() for kw in tech_keywords)

    async def _detect_tech_stack(self, question: str) -> Optional[str]:
        """检测技术栈"""
        prompt = f"""从以下问题中检测涉及的技术栈，只返回技术名称：
        问题：{question}
        技术栈（如 python、react、docker 等）："""
        response = await self.llm.chat([
            Message(role="user", content=prompt)
        ])
        tech = response.strip().lower()
        tech_docs = {
            "python": "docs.python.org",
            "javascript": "developer.mozilla.org",
            "react": "react.dev",
            "vue": "vuejs.org",
            "docker": "docs.docker.com",
            "kubernetes": "kubernetes.io",
        }
        return tech_docs.get(tech)

    def _get_docs_url(self, tech: str) -> str:
        """获取文档站点 URL"""
        tech_docs = {
            "python": "docs.python.org",
            "javascript": "developer.mozilla.org",
            "react": "react.dev",
            "vue": "vuejs.org",
            "docker": "docs.docker.com",
        }
        return tech_docs.get(tech, "docs.python.org")

    async def _fetch_page_contents(self, results: List[SearchResult]) -> List[SearchResult]:
        """获取页面完整内容"""
        async def fetch_content(result: SearchResult):
            try:
                async with aiohttp.ClientSession() as session:
                    async with session.get(
                        result.url, headers=self.search.headers, timeout=aiohttp.ClientTimeout(total=10)
                    ) as response:
                        html = await response.text()
                        from bs4 import BeautifulSoup
                        soup = BeautifulSoup(html, 'html.parser')
                        for script in soup(['script', 'style', 'nav', 'footer']):
                            script.decompose()
                        text = soup.get_text(separator='\n', strip=True)
                        result.snippet = text[:2000] + "..."
                        result.relevance = 1.0
            except Exception as e:
                print(f" ⚠️ 获取失败 {result.url}: {e}")
        
        tasks = [fetch_content(r) for r in results]
        await asyncio.gather(*tasks)
        return results

    async def _synthesize_answer(self, question: str, results: List[SearchResult]) -> str:
        """综合搜索结果生成答案"""
        context = "\n\n".join([
            f"来源{i+1}: {r.title}\n{r.snippet}\n链接：{r.url}"
            for i, r in enumerate(results[:5])
        ])
        prompt = f"""基于以下搜索结果回答问题，要求：
        1. 准确引用信息来源
        2. 综合多个来源的信息
        3. 如果信息冲突，说明不同观点
        4. 给出清晰的结构化答案
        5. 标注信息来源（如 [来源 1]）
        问题：{question}
        搜索结果：
        {context}
        请给出详细答案："""
        answer = await self.llm.chat([
            Message(role="system", content="你是一个专业的研究助手，擅长综合多源信息给出准确答案"),
            Message(role="user", content=prompt)
        ])
        return answer

    async def _generate_related_questions(self, question: str, answer: str) -> List[str]:
        """生成相关问题"""
        prompt = f"""基于以下问答，生成 3-5 个相关的深入研究问题：
        问题：{question}
        答案：{answer[:500]}...
        请生成相关问题，每行一个："""
        response = await self.llm.chat([
            Message(role="user", content=prompt)
        ])
        return [
            line.strip()
            for line in response.split('\n')
            if line.strip() and not line.startswith('-')
        ][:5]

    def _calculate_confidence(self, results: List[SearchResult]) -> float:
        """计算答案置信度"""
        if not results:
            return 0.0
        base_confidence = min(1.0, len(results)/10)
        has_docs = any(r.source == "docs" for r in results)
        if has_docs:
            base_confidence = min(1.0, base_confidence + 0.2)
        return round(base_confidence, 2)

# 使用示例
async def main_researcher():
    llm = LLMClient()
    search = SearchEngine()
    researcher = IntelligentResearcher(llm, search)
    
    result = await researcher.research(
        question="Python 中 asyncio 和 multiprocessing 的区别是什么？",
        depth=2
    )
    print("\n" + "="*60)
    print("📚 研究结果")
    print("="*60)
    print(f"\n置信度：{result.confidence*100}%\n")
    print(f"答案:\n{result.answer}\n")
    print("📖 参考来源:")
    for i, source in enumerate(result.sources, 1):
        print(f"{i}. {source.title}")
        print(f" 链接：{source.url}")
        print(f" 来源：{source.source}\n")
    print("❓ 相关问题:")
    for q in result.related_questions:
        print(f" • {q}")

if __name__ == "__main__":
    asyncio.run(main_researcher())

4.3 搜索效率对比

五、整合三大利器：打造超级 AI 助手

5.1 统一 CLI 工具

我们将三个工具打包成一个命令行工具，方便集成到工作流中。

import argparse
import asyncio
from pathlib import Path
import json

class AIToolsCLI:
    """AI 工具命令行界面"""
    def __init__(self):
        self.llm = LLMClient()
        self.summarizer = DocumentSummarizer(self.llm)
        self.code_assistant = InteractiveCodeAssistant(self.llm)
        self.researcher = IntelligentResearcher(
            self.llm, SearchEngine()
        )

    async def run(self):
        """运行 CLI"""
        parser = argparse.ArgumentParser(
            description="AI 工具集 - 你的智能助手",
            formatter_class=argparse.RawDescriptionHelpFormatter,
            epilog="""
            示例:
            # 总结文档
            python ai_tools.py summarize paper.pdf
            # 生成代码
            python ai_tools.py code "用 Python 写一个爬虫"
            # 研究问题
            python ai_tools.py research "量子计算的原理"
            """
        )
        subparsers = parser.add_subparsers(dest='command', help='可用命令')
        
        # summarize 命令
        sum_parser = subparsers.add_parser('summarize', help='总结文档')
        sum_parser.add_argument('file', help='文件路径或 URL')
        sum_parser.add_argument('-t', '--type', default='file', choices=['file', 'url'], help='输入类型')
        sum_parser.add_argument('-o', '--output', help='输出文件路径')
        
        # code 命令
        code_parser = subparsers.add_parser('code', help='生成/处理代码')
        code_parser.add_argument('prompt', help='需求或代码')
        code_parser.add_argument('-m', '--mode', choices=['generate', 'explain', 'optimize', 'debug'], default='generate', help='处理模式')
        code_parser.add_argument('-l', '--language', default='python', help='编程语言')
        code_parser.add_argument('-x', '--execute', action='store_true', help='执行生成的代码')
        
        # research 命令
        res_parser = subparsers.add_parser('research', help='研究问题')
        res_parser.add_argument('question', help='研究问题')
        res_parser.add_argument('-d', '--depth', type=int, default=1, choices=[1, 2, 3], help='研究深度')
        res_parser.add_argument('-s', '--sources', nargs='+', choices=['google', 'docs', 'stackoverflow'], help='指定搜索源')
        
        args = parser.parse_args()
        if not args.command:
            parser.print_help()
            return
        
        if args.command == 'summarize':
            await self._cmd_summarize(args)
        elif args.command == 'code':
            await self._cmd_code(args)
        elif args.command == 'research':
            await self._cmd_research(args)

    async def _cmd_summarize(self, args):
        """处理 summarize 命令"""
        print(f"📖 正在总结：{args.file}")
        result = await self.summarizer.summarize(
            source=args.file, source_type=args.type
        )
        output = f"""# {result.title}
        **📊 统计信息**
        - 字数：{result.word_count}
        - 预计阅读时间：{result.reading_time} 分钟
        - 生成时间：{result.created_at}
        **🔑 关键要点**
        {chr(10).join(f'{i+1}. {p}' for i, p in enumerate(result.key_points))}
        **📝 总结**
        {result.summary}
        """
        if args.output:
            with open(args.output, 'w', encoding='utf-8') as f:
                f.write(output)
            print(f"✅ 已保存到：{args.output}")
        else:
            print(output)

    async def _cmd_code(self, args):
        """处理 code 命令"""
        print(f"💻 正在处理：{args.prompt[:50]}...")
        result = await self.code_assistant.generator.generate(
            requirement=args.prompt, language=args.language, mode=CodeMode(args.mode)
        )
        print(f"\n```{args.language}")
        print(result.code)
        print("```\n")
        print(f"**说明**\n{result.explanation}\n")
        if result.warnings:
            print("**警告**")
            for w in result.warnings:
                print(f" {w}")
            print()
        if result.tests:
            print(f"**测试代码**\n```{args.language}")
            print(result.tests)
            print("```\n")
        if args.execute:
            print("⚡ 正在执行代码...")
            exec_result = await self.code_assistant.generator.execute_code(
                result.code, args.language
            )
            if exec_result['success']:
                print(f"✅ 执行成功\n输出:\n{exec_result['output']}")
            else:
                print(f"❌ 执行失败\n错误:\n{exec_result['error']}")

    async def _cmd_research(self, args):
        """处理 research 命令"""
        print(f"🔍 正在研究：{args.question}")
        result = await self.researcher.research(
            question=args.question, depth=args.depth, sources=args.sources
        )
        print(f"""
        # 研究结果
        **📊 置信度**: {result.confidence*100}%
        ## 答案
        {result.answer}
        ## 参考来源
        """)
        for i, source in enumerate(result.sources, 1):
            print(f"{i}. **{source.title}**")
            print(f" 链接：{source.url}")
            print(f" 来源：{source.source}\n")
        if result.related_questions:
            print("## 相关问题")
            for q in result.related_questions:
                print(f"- {q}")

async def main():
    cli = AIToolsCLI()
    await cli.run()

if __name__ == "__main__":
    asyncio.run(main())

5.2 使用示例

# 总结论文
python ai_tools.py summarize research_paper.pdf -o summary.md

# 生成代码并执行
python ai_tools.py code "用 Python 写一个二分查找" -x

# 解释代码
python ai_tools.py code "explain this code: `def foo(): return 1`" -m explain

# 深度研究
python ai_tools.py research "RAG 和 Fine-tuning 的区别" -d 2

5.3 成本分析

六、完整源码与部署指南

6.1 项目结构

ai-tools/
├── src/
│   ├── __init__.py
│   ├── llm.py          # LLM 客户端
│   ├── summarizer.py   # 文档总结器
│   ├── code_generator.py # 代码生成器
│   └── researcher.py   # 研究助手
├── cli.py              # 命令行入口
├── config.py           # 配置管理
├── requirements.txt    # 依赖列表
├── .env.example        # 环境变量示例
├── README.md           # 使用文档
└── examples/           # 使用示例
    ├── example_summarize.py
    ├── example_code.py
    └── example_research.py

6.2 部署到云端

FROM python:3.11-slim
WORKDIR /app
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt
COPY src/ ./src/
COPY cli.py .
COPY config.py .
ENV PYTHONPATH=/app
CMD ["python", "cli.py", "--help"]

# docker-compose.yml
version: '3.8'
services:
  ai-tools:
    build: .
    env_file: - .env
    volumes:
      - ./data:/app/data
    ports:
      - "8000:8000"

6.3 进阶功能扩展

总结

通过这篇文章，我们用 Python 打造了三个强大的 AI 工具：

关键收获

1. LLM 调用很简单 - 用好 OpenAI SDK，30 行代码就能连接大模型
2. 提示词是关键 - 好的 prompt 能让效果翻倍
3. 异步处理很重要 - 并行请求能大幅提升速度
4. 安全意识不能少 - 代码执行要隔离，API 调用要限流

下一步学习

• 📖 深入学习 LangChain 框架
• 🔬 研究 Agent 与 RAG 技术
• 🚀 打造专属 AI 应用

参考资源

• OpenAI Python SDK 文档
• DeepSeek API 文档
• 硅基流动平台
• Python 异步编程指南