CareerBot/app/services/chat_service.py

from datetime import datetime
from typing import AsyncGenerator

from fastapi import UploadFile
from sqlalchemy.orm import Session

from app.models import Profile, Skill, Education, WorkExperience, ChatHistory, LLMConfig
from app.services import llm_service
from app.services.file_parser import parse_file


def build_system_prompt(db: Session) -> str:
    """Build system prompt from database content."""
    profile = db.query(Profile).first()
    skills = db.query(Skill).order_by(Skill.sort_order).all()
    educations = db.query(Education).order_by(Education.sort_order).all()
    experiences = db.query(WorkExperience).order_by(WorkExperience.sort_order).all()

    # Check for custom system prompt in LLM config
    config = db.query(LLMConfig).filter(LLMConfig.is_active == True).first()
    custom_prompt = config.system_prompt if config and config.system_prompt else ""

    name = profile.name if profile else "候选人"

    parts = [
        f"你是{name}的个人AI职业助手，专门帮助招聘者了解{name}的职业背景、技能和工作经验。",
        "",
    ]

    if profile:
        parts.append("=== 个人基本信息 ===")
        parts.append(f"姓名: {profile.name}")
        parts.append(f"学历: {profile.education_level}")
        parts.append(f"所在地: {profile.location}")
        parts.append(f"邮箱: {profile.email}")
        if profile.self_summary:
            parts.append(f"个人总结: {profile.self_summary}")
        parts.append("")

    if skills:
        parts.append("=== 技能特长 ===")
        for s in skills:
            parts.append(f"- {s.category}: {s.content}")
        parts.append("")

    if educations:
        parts.append("=== 教育经历 ===")
        for e in educations:
            parts.append(f"{e.start_date} - {e.end_date} | {e.school} | {e.major} | {e.degree}")
            if e.details:
                parts.append(f"  详情: {e.details}")
        parts.append("")

    if experiences:
        parts.append("=== 工作经历 ===")
        for exp in experiences:
            parts.append(f"{exp.start_date} - {exp.end_date} | {exp.company} | {exp.position}")
            if exp.company_intro:
                parts.append(f"  公司简介: {exp.company_intro}")
            if exp.responsibilities:
                parts.append(f"  工作职责: {exp.responsibilities}")
            if exp.achievements:
                parts.append(f"  工作成就: {exp.achievements}")
            parts.append("")

    parts.append("=== 回答规则 ===")
    parts.append("1. 只回答与招聘、职业、工作能力、技术背景相关的问题")
    parts.append("2. 对于无关问题，礼貌拒绝并引导回职业话题")
    parts.append("3. 严格基于以上信息回答，不编造经历")
    parts.append("4. 使用专业、友好的语气")
    parts.append("5. 如果招聘者上传了职位描述(JD)，详细分析岗位匹配度，突出亮点、优势和相关经历，同时客观指出可能的不足")
    parts.append("6. 根据招聘者使用的语言选择中文或英文回答")

    if custom_prompt:
        parts.append("")
        parts.append("=== 额外指令 ===")
        parts.append(custom_prompt)

    return "\n".join(parts)


async def process_message(
    session_id: str,
    user_message: str,
    db: Session,
    file: UploadFile | None = None,
) -> AsyncGenerator[str, None]:
    """Process a chat message and stream the response."""
    # Handle file upload
    file_context = ""
    image_data = None
    if file and file.filename:
        parsed = await parse_file(file)
        if parsed.get("error"):
            file_context = f"\n[文件解析提示: {parsed['error']}]"
        elif parsed.get("text"):
            file_context = f"\n\n招聘者上传的职位描述(JD)内容如下:\n{parsed['text']}\n\n请分析此岗位与我的匹配度，包括匹配的方面、亮点优势和可能的不足。"
        if parsed.get("base64_image"):
            image_data = {
                "base64": parsed["base64_image"],
                "mime_type": parsed.get("mime_type", "image/png"),
            }

    full_user_message = user_message + file_context

    # Get chat history
    history = (
        db.query(ChatHistory)
        .filter(ChatHistory.session_id == session_id)
        .order_by(ChatHistory.created_at.desc())
        .limit(20)
        .all()
    )
    history.reverse()

    # Build messages
    system_prompt = build_system_prompt(db)
    messages = [{"role": "system", "content": system_prompt}]

    for h in history:
        messages.append({"role": h.role, "content": h.content})

    # Format user message (with optional image)
    if image_data:
        user_content = [
            {"type": "text", "text": full_user_message or "请分析这个职位描述图片与我的匹配度。"},
            {
                "type": "image_url",
                "image_url": {
                    "url": f"data:{image_data['mime_type']};base64,{image_data['base64']}"
                },
            },
        ]
        messages.append({"role": "user", "content": user_content})
    else:
        messages.append({"role": "user", "content": full_user_message})

    # Save user message to history
    db.add(ChatHistory(
        session_id=session_id,
        role="user",
        content=full_user_message,
        created_at=datetime.utcnow(),
    ))
    db.commit()

    # Stream response
    full_response = []
    async for chunk in llm_service.chat_completion_stream(messages, db):
        full_response.append(chunk)
        yield chunk

    # Save assistant response (skip error messages)
    assistant_content = "".join(full_response)
    if assistant_content and "AI服务调用失败" not in assistant_content:
        db.add(ChatHistory(
            session_id=session_id,
            role="assistant",
            content=assistant_content,
            created_at=datetime.utcnow(),
        ))
        db.commit()