zhanghongyu_0626
4b8e11c9bd
- Added `ThresholdRepository` for managing customer loss threshold configurations and notifications. - Introduced `threshold_service` to handle loss threshold alerts based on customer portfolio performance. - Enhanced `customer_prompts` to include new intent for querying product net values. - Updated `customer_service` to integrate new threshold alert functionality into existing workflows. - Implemented `sanitize_postprocess` for improved compliance handling in customer interactions. - Enhanced course documentation to reflect updates in advisor training modules and interactive elements. This update significantly improves the customer experience by providing proactive loss threshold notifications and enhancing the overall service framework.
147 lines
8.5 KiB
HTML
147 lines
8.5 KiB
HTML
<section class="module" id="module-2">
|
||
<div class="module-inner">
|
||
<p class="eyebrow animate-in">模块 2 · 三张牌编排</p>
|
||
<h1 class="module-title animate-in">tool → llm → guard<br>群聊走一遍</h1>
|
||
<p class="module-lead animate-in">
|
||
顾问线 LangGraph 只有三个业务节点（加 START/END）：
|
||
<span class="term" data-definition="tool 节点 = 根据用户消息关键词匹配 Tool 名，调 对话 Tool 编排（tool_service.py）run_tool 查库并落 agent_tool_call 留痕。">tool</span>
|
||
→ llm → guard。比客户 14 节点图短，但<strong>归属校验在 对话 Tool 编排（tool_service.py）run_tool 里</strong>，和客户线铁律口径一致。
|
||
</p>
|
||
|
||
<div class="screen animate-in">
|
||
<h2>节点职责（背下来少踩坑）</h2>
|
||
<div class="pattern-cards">
|
||
<div class="pattern-card">
|
||
<h3>tool_node</h3>
|
||
<p><code>match_intent("advisor", message)</code> 命中关键词 → 对话 Tool 编排（<code>tool_service.py</code>）<code>run_tool</code> 查 Core / 知识库 → 结果写入 <code>tool_results</code>。</p>
|
||
</div>
|
||
<div class="pattern-card">
|
||
<h3>llm_node</h3>
|
||
<p>System prompt（顾问边界）+ Tool 摘要注入 + 历史窗口 + 本轮用户话 → DeepSeek 组织自然语言。</p>
|
||
</div>
|
||
<div class="pattern-card">
|
||
<h3>guard_node</h3>
|
||
<p>顾问线<strong>不附</strong>客户/风控那种尾部免责声明（<code>needs_disclaimer("advisor")</code> 为 false）。</p>
|
||
</div>
|
||
</div>
|
||
</div>
|
||
|
||
<div class="screen animate-in">
|
||
<h2>群聊：理财师问「客户 CUST-9527 持仓如何」</h2>
|
||
<p>简化播放 tool → llm → guard 全链路（前面 四 Agent 对话 HTTP 入口（chat.py）已验 JWT + G-01）。</p>
|
||
|
||
<div class="chat-window" id="chat-advisor-m2">
|
||
<div class="chat-messages">
|
||
<div class="chat-message" data-msg="0" data-sender="browser" style="display:none">
|
||
<div class="chat-avatar" style="background:#D4A843">前</div>
|
||
<div class="chat-bubble">
|
||
<span class="chat-sender" style="color:#D4A843">浏览器</span>
|
||
<p>POST /api/chat/stream · X-Agent-Type: advisor · customer_id: CUST-9527</p>
|
||
</div>
|
||
</div>
|
||
<div class="chat-message" data-msg="1" data-sender="chat" style="display:none">
|
||
<div class="chat-avatar" style="background:#B8922F">API</div>
|
||
<div class="chat-bubble">
|
||
<span class="chat-sender" style="color:#B8922F">四 Agent 对话 HTTP 入口（chat.py）</span>
|
||
<p>验 JWT → assert_customer_access（G-01：名下客户）→ 建/续 session</p>
|
||
</div>
|
||
</div>
|
||
<div class="chat-message" data-msg="2" data-sender="tool" style="display:none">
|
||
<div class="chat-avatar" style="background:#E0C06A">Tool</div>
|
||
<div class="chat-bubble">
|
||
<span class="chat-sender" style="color:#E0C06A">tool_node</span>
|
||
<p>关键词「持仓」→ <code>query_holdings</code> → 对话 Tool 编排（tool_service.py）run_tool</p>
|
||
</div>
|
||
</div>
|
||
<div class="chat-message" data-msg="3" data-sender="core" style="display:none">
|
||
<div class="chat-avatar" style="background:#2D8B55">DB</div>
|
||
<div class="chat-bubble">
|
||
<span class="chat-sender" style="color:#2D8B55">core_ro</span>
|
||
<p>只读 SQL 查持仓 → fact_text 摘要写入 tool_results；agent_tool_call 落库 success</p>
|
||
</div>
|
||
</div>
|
||
<div class="chat-message" data-msg="4" data-sender="llm" style="display:none">
|
||
<div class="chat-avatar" style="background:#7B6DAA">AI</div>
|
||
<div class="chat-bubble">
|
||
<span class="chat-sender" style="color:#7B6DAA">llm_node</span>
|
||
<p>Tool 摘要注入 system 上下文 → DeepSeek 生成顾问口吻回复（数据引用 fact_text）</p>
|
||
</div>
|
||
</div>
|
||
<div class="chat-message" data-msg="5" data-sender="guard" style="display:none">
|
||
<div class="chat-avatar" style="background:#2A7B9B">盾</div>
|
||
<div class="chat-bubble">
|
||
<span class="chat-sender" style="color:#2A7B9B">guard_node</span>
|
||
<p>advisor 不拼 CHAT_DISCLAIMER；reply 原样传出</p>
|
||
</div>
|
||
</div>
|
||
<div class="chat-message" data-msg="6" data-sender="sse" style="display:none">
|
||
<div class="chat-avatar" style="background:#D4A843">流</div>
|
||
<div class="chat-bubble">
|
||
<span class="chat-sender" style="color:#D4A843">stream_chat</span>
|
||
<p>Tool 同步跑完后逐块推 LLM 文本；收完 done 后 api 层统一落 MySQL 回合</p>
|
||
</div>
|
||
</div>
|
||
</div>
|
||
<div class="chat-typing" id="chat-advisor-m2-typing" style="display:none">
|
||
<div class="chat-avatar" id="chat-advisor-m2-typing-avatar">…</div>
|
||
<div class="chat-typing-dots"><span class="typing-dot"></span><span class="typing-dot"></span><span class="typing-dot"></span></div>
|
||
</div>
|
||
<div class="chat-controls">
|
||
<button class="btn chat-next-btn">下一条</button>
|
||
<button class="btn chat-all-btn">全部播放</button>
|
||
<button class="btn chat-reset-btn">重播</button>
|
||
<span class="chat-progress"></span>
|
||
</div>
|
||
</div>
|
||
</div>
|
||
|
||
<div class="screen animate-in">
|
||
<div class="translation-block">
|
||
<div class="translation-code">
|
||
<span class="translation-label">CODE · 顾问通用 tool→llm→guard 编排（agent_service.py）tool_node</span>
|
||
<pre><code><span class="code-line">tool_name = tool_service.match_intent(</span>
|
||
<span class="code-line"> state[<span class="code-string">"agent_type"</span>], state[<span class="code-string">"user_message"</span>]</span>
|
||
<span class="code-line">)</span>
|
||
<span class="code-line">record = tool_service.run_tool(</span>
|
||
<span class="code-line"> tool_name=tool_name,</span>
|
||
<span class="code-line"> agent_type=state[<span class="code-string">"agent_type"</span>],</span>
|
||
<span class="code-line"> customer_id=state.get(<span class="code-string">"customer_id"</span>) or <span class="code-string">""</span>,</span>
|
||
<span class="code-line"> ...</span>
|
||
<span class="code-line">)</span></code></pre>
|
||
</div>
|
||
<div class="translation-english">
|
||
<span class="translation-label">白话</span>
|
||
<div class="translation-lines">
|
||
<p class="tl">先靠关键词判断要不要查库；闲聊不命中就 tool_results 为空，LLM 纯聊。</p>
|
||
<p class="tl">customer_id 来自会话绑定（四 Agent 对话 HTTP 入口（chat.py）归属校验后的值），不是 LLM 生成的。</p>
|
||
<p class="tl">query_holdings 和 customer 线共用 Core 只读 Tool 定义（core_tools.py）定义，但跑在 顾问通用 tool→llm→guard 编排（agent_service.py）图里。</p>
|
||
</div>
|
||
</div>
|
||
</div>
|
||
|
||
<div class="quiz-container" id="quiz-advisor-m2">
|
||
<div class="quiz-question-block"
|
||
data-correct="option-a"
|
||
data-explanation-right="对。一期用关键词 match_intent；命中「持仓」等词才调 query_holdings。"
|
||
data-explanation-wrong="顾问线 Tool 意图是一期关键词规则，不是 LLM 自由选函数名。">
|
||
<h3 class="quiz-question">理财师问「这位客户仓位怎么样」，tool_node 如何选中 query_holdings？</h3>
|
||
<div class="quiz-options">
|
||
<button class="quiz-option" data-value="option-a" onclick="selectOption(this)">
|
||
<div class="quiz-option-radio"></div><span>match_intent 关键词命中「仓位」</span>
|
||
</button>
|
||
<button class="quiz-option" data-value="option-b" onclick="selectOption(this)">
|
||
<div class="quiz-option-radio"></div><span>LLM 在回复里输出函数名</span>
|
||
</button>
|
||
<button class="quiz-option" data-value="option-c" onclick="selectOption(this)">
|
||
<div class="quiz-option-radio"></div><span>前端在 body 里传 tool_name</span>
|
||
</button>
|
||
</div>
|
||
<div class="quiz-feedback"></div>
|
||
</div>
|
||
<button class="quiz-check-btn" onclick="checkQuiz('quiz-advisor-m2')">检查答案</button>
|
||
<button class="quiz-reset-btn" onclick="resetQuiz('quiz-advisor-m2')">重做</button>
|
||
</div>
|
||
</div>
|
||
</div>
|
||
</section>
|