意图路由
对传入的请求进行分类，并把每一个路由到最合适的处理方：确定性逻辑、专用 LLM，或人工。
并非每个用户请求都需要同一种处理方。有些可以通过一次数据库查询来回答。有些需要一个具备领域特定上下文的 LLM。有些需要人工。TypeSafe 可以作为快速、廉价的分类器置于所有这些处理方之前，判定应当调用哪一个。
示例：客户服务路由
设想你正在构建一个客户服务系统。消息不断涌入，需要被路由到正确的处理方。与其把每条消息都送进一个昂贵的 LLM 来判断它属于哪类请求，不如先做分类，再据此路由。
%%{init: {"fontFamily": "Inter, sans-serif", "flowchart": {"rankSpacing": 35, "wrappingWidth": 300, "subGraphTitleMargin": {"top": 12, "bottom": 36}}}}%%
flowchart LR
message["客户消息"]
subgraph req["TypeSafe 并行<br/>评估各个问题"]
direction TB
intent["<b>Choice:</b> 意图"]
complexity["<b>Score:</b> 复杂度"]
%% Invisible links stack the questions; they are answered in parallel.
intent ~~~ complexity
end
message -- "一次请求<br/>消息 + 2 个问题" --> req
req -- "一次响应<br/>2 个答案<br/>及置信度" --> confidence{"<b>意图置信度<br/>≥ 0.5？</b><br/>你的代码"}
confidence -- "否" --> human["人工客服"]
confidence -- "是" --> route{"<b>是哪个意图？</b><br/>"}
route -- "order_status" --> order["订单查询<br/>确定性代码"]
route -- "product_question" --> product["产品专家 LLM"]
route -- "return_exchange" --> returns["退货专家 LLM"]
route -- "complaint" --> escalate{"<b>复杂度 > 1<br/>或其置信度 < 0.5？</b><br/>"}
escalate -- "是" --> human
escalate -- "否" --> complaint["投诉解决 LLM"]
第 1 步：对意图与复杂度进行分类
{
"questions": {
"intent": {
"type": "choice",
"instructions": "The primary intent of this customer message",
"criteria": {
"order_status": "Asking about an existing order",
"product_question": "Asking about a product before buying",
"return_exchange": "Wants to return or exchange something",
"complaint": "Unhappy with experience, wants resolution"
}
},
"complexity": {
"type": "score",
"instructions": "How complex is this request to resolve",
"criteria": [
"Simple lookup or standard procedure",
"Requires some judgment or multi-step process",
"Unusual situation, edge case, or escalation needed"
]
}
}
}这个示例是可交互的；到官网原页面可以直接在 Playground 里运行。
第 2 步：路由到最合适的处理方
def route_ticket(ticket_id, response):
intent = response.answers["intent"]
complexity = response.answers["complexity"]
if intent.confidence < 0.5:
# If we don't have enough confidence to classify, route to a human agent
return route_to_human_agent(ticket_id)
if intent.choice == "order_status":
handle_order_status(ticket_id)
elif intent.choice == "product_question":
handle_with_llm(ticket_id, PRODUCT_SPECIALIST)
elif intent.choice == "return_exchange":
handle_with_llm(ticket_id, RETURNS_SPECIALIST)
elif intent.choice == "complaint":
low_confidence = complexity.confidence < 0.5
# A higher complexity.score leans toward the "escalation needed" end of the scale.
if complexity.score > 1 or low_confidence:
# Too complex for safe automation, or we're not sure about the complexity; route to a human.
route_to_human_agent(ticket_id)
else:
handle_with_llm(ticket_id, COMPLAINT_RESOLUTION)
有一种意图被路由到不涉及任何 LLM 的确定性代码。有两种被路由到不同的专用 LLM，各自加载不同的上下文。还有一种利用复杂度分数在 LLM 与人工之间做出选择。TypeSafe 在单次快速调用中完成全部分类；昂贵的资源只会为真正需要它们的请求而启用。
注意对复杂度分数所做的额外置信度检查。正如置信度中所讨论的，在系统所处的语境与决策的代价这一背景下，考虑低置信度分数的含义始终很重要。