接听真实电话的 AI 接待员
1 分•作者: kaansarac•8 个月前
我们正在构建一个 AI 接待员,可以接听真实的电话,捕捉潜在客户,预约,并发送 5 分钟的跟进邮件。我们的第一个细分市场是婚礼场地。我是其中一位创始人。
对 HN 可能有意思的地方:
对话循环:电话 → 流式 ASR(自动语音识别)→ LLM(大型语言模型)工具 → 日历/电子邮件 → TTS(文本转语音),带有轮流对话和抢先控制。
日历预约:缓冲时间逻辑 + 防止重复预订;我们公开一个最小的功能 API,用于“提供时段/预订时段/确认”。
多渠道捕获:将电话、电子邮件和表单线索统一到一个记录中,包括文字记录 + 字段(姓名、日期、宾客人数、预算)。
垃圾邮件过滤:在垃圾邮件(例如,1-800 电话/机器人电话)到达工作人员之前进行拦截;VIP 安全直通规则。
评估工具:脚本化通话场景(可用性、定价、政策)→ 检查依据(答案必须在您的文档中)→ 针对正确性、安全性、和升级时机进行评分。
哪些行不通:
在知识摄取之前就急于回答;我们现在对来自已验证来源的答案进行硬性限制,否则会留言或升级。
Elevenlabs;延迟太高,无法构建类似人类的体验。
对边缘情况的困惑(“如果…,您的取消政策是什么”)。我们添加了文档优先检索 + 回退到“收集信息 + 路由”。
目前的数据(早期,仅有 6 个客户,并且正在改进):
目标应答时间:小于一秒的接听;电子邮件 5 分钟内首次回复。
减少未接电话和更快的行程安排是主要优势;在数据成熟后很乐意分享更多信息。
隐私/伦理:
客户内容不会用于训练我们的模型。
明确的同意和录音政策;PII(个人身份信息)在静态和传输过程中均已加密。
我希望获得的反馈:
更好的语音代理离线评估(超越正常路径脚本)。
您认为有效的轮流对话和抢先策略。
在信任 AI 接听电话之前,您希望看到的故障模式处理。
链接:https://mikla.ai
查看原文
We’re building an AI receptionist that answers real phone calls, captures leads, books appointments, and sends 5‑minute follow‑ups. Our first niche is wedding venues. I’m one of the founders.<p>What might be interesting to HN:<p>Conversation loop: telephony → streaming ASR → LLM tools → calendar/email → TTS, with turn‑taking and barge‑in control.
Calendar booking: buffer‑time logic + double‑booking prevention; we expose a minimal function API for “OfferSlots/BookSlot/Confirm.”
Multi‑channel capture: unify phone, email, and form leads into one record with transcript + fields (name, date, guest count, budget).
Spam filtering: block patterns (e.g., 1‑800s / robocalls) before they hit staff; safe pass‑through rules for VIPs.
Evaluation harness: scripted call scenarios (availability, pricing, policy) → check for grounding (answers must be in your docs) → score for correctness, safety, and escalation timing.<p>What didn’t work:<p>Over‑eager answers before knowledge ingestion; we now hard‑gate answers on verified sources and otherwise take a message or escalate.
Elevenlabs; Latency is way too much to build a human like experience.
Confusion on edge cases (“What’s your cancellation policy if…”). We added doc‑first retrieval + fallback to “collect info + route.”<p>Numbers so far (early, only 6 customers and improving):<p>Target answer time: sub‑second pickup; 5‑minute first reply on email.
Reduction in missed calls and faster tour scheduling are the main wins; happy to share more once data matures.<p>Privacy/ethics:<p>Customer content is not used to train our models.
Clear consent and recording policies; PII is encrypted at rest and in transit.
What I’d love feedback on:<p>Better offline evaluation for voice agents (beyond happy‑path scripts).
Turn‑taking and barge‑in strategies you’ve found to work well.
Failure‑mode handling you’d want before trusting an AI with calls.<p>Link: https://mikla.ai