Skip to content

Project: Multilingual Customer Support Assistant

TR/EN support pipeline: intent, FAQ, escalation, CSAT feedback. Which step on which model, and how to design evals.

Şükrü Yusuf KAYA
14 min read
Advanced
Support bot mimarisi: intent → FAQ → eskalasyon

Pipeline Tasarımı

Customer message
   ↓
[1] Language detect (Haiku)
   ↓
[2] Intent classify (Haiku)
   ↓
[3] FAQ retrieval (vector DB)  ←  yeterli mi?
   ↓                              ↑
   ├─ Evet → answer (Sonnet)
   └─ Hayır → escalate to human + draft a suggested reply (Sonnet)
   ↓
[4] CSAT feedback collect → eval set
Maliyet: ortalama mesaj başına 0.001(Haiku)+ 0.001 (Haiku) + ~0.005 (Sonnet) = **$0.006**.
python
# Üst düzey iskelet
def support_pipeline(msg, user):
    lang   = haiku("dili belirle: {msg}")
    intent = haiku(intent_prompt(msg, lang), schema=INTENT_SCHEMA)
    if intent["confidence"] < 0.6:
        return escalate(msg, user, reason="low_confidence_intent")
 
    chunks = vector_search(msg, k=5, lang=lang)
    if not chunks or top_score(chunks) < 0.7:
        return escalate(msg, user, reason="no_faq_match",
                        suggested=draft_reply(msg, intent))
 
    return sonnet(answer_prompt(msg, lang, chunks), citations=True)
Yüksek seviye iskelet — gerçek kodda her çağrı log + retry + cache.
Boşluk doldur · text
Support pipeline'ında intent confidence eşiği genelde _____'tir. FAQ retrieval skoru _____ altındaysa eskalasyon tetiklenir. Geri besleme döngüsünde _____ skoru kullanılır.

Frequently Asked Questions

Best practice: ingest the human resolution back into your knowledge base. Over time, escalation rates drop.

Yorumlar & Soru-Cevap

(0)
Yorum yazmak için giriş yap.
Yorumlar yükleniyor...

Related Content

11. Real-World Projects

Project: RAG Document Q&A System

Start Learning
11. Real-World Projects

Project: Code Review Automation

Start Learning
1. Foundations — Welcome to Claude

What is Claude? The New Generation of AI Assistants

Start Learning

Connected pillar topics

Pillar topics this article maps to

Pillar Topic

RAG (Retrieval-Augmented Generation) Architecture

RAG (Retrieval-Augmented Generation) is an architecture that grounds large-language-model answers in chunks retrieved from the organization's own documents or data sources, providing both freshness and citations.