AI Chatbot Examples

A chatbot that handles tier-1 support queries using a knowledge base of FAQs and product documentation. When confidence drops below a threshold or the customer expresses frustration, it seamlessly hands off to a human agent with full conversation context.

// Escalation logic pseudocode
async function handleMessage(message, context) {
  const intent = await classifyIntent(message);
  const confidence = intent.confidence;
  const sentiment = await analyseSentiment(message);

  if (confidence < 0.6 || sentiment.score < -0.5 || context.turnCount > 5) {
    return escalateToHuman(context);
  }

  const response = await generateResponse(intent, context.knowledgeBase);
  return { reply: response, escalate: false };
}

An AI assistant integrated with ServiceNow that helps employees reset passwords, check ticket status, and troubleshoot common issues. Uses RAG over internal documentation to answer policy questions.

// Tool-use pattern for IT helpdesk
const tools = [
  { name: "resetPassword", params: ["employeeId"] },
  { name: "checkTicketStatus", params: ["ticketId"] },
  { name: "createTicket", params: ["category", "description", "priority"] },
  { name: "searchKnowledgeBase", params: ["query"] },
];

// The LLM decides which tool to invoke based on user intent
const toolCall = await llm.chat({
  messages: conversationHistory,
  tools: tools,
  systemPrompt: "You are an IT helpdesk assistant..."
});

A website chatbot that engages visitors, qualifies them against BANT criteria (Budget, Authority, Need, Timeline), and books meetings for sales reps. Integrates with CRM to log interactions and scoring.

// BANT qualification flow
const qualificationCriteria = {
  budget: { question: "What budget range are you considering?", weight: 0.3 },
  authority: { question: "Who else is involved in this decision?", weight: 0.25 },
  need: { question: "What problem are you trying to solve?", weight: 0.25 },
  timeline: { question: "When are you looking to implement?", weight: 0.2 },
};

function calculateLeadScore(responses) {
  return Object.entries(responses).reduce((score, [key, value]) => {
    return score + (qualificationCriteria[key].weight * value.score);
  }, 0);
}

A chatbot serving customers across 12 languages by detecting the input language, translating to English for processing, generating a response, then translating back. Uses a single English knowledge base to avoid maintaining parallel content.

// Multi-lingual pipeline
async function processMultiLingual(message) {
  const detectedLang = await detectLanguage(message);

  // Translate to canonical language for processing
  const englishMessage = detectedLang !== "en"
    ? await translate(message, detectedLang, "en")
    : message;

  // Process against English knowledge base
  const response = await generateResponse(englishMessage);

  // Translate response back to user's language
  return detectedLang !== "en"
    ? await translate(response, "en", detectedLang)
    : response;
}

A phone-based AI receptionist using speech-to-text and text-to-speech that answers calls, routes enquiries, takes messages, and schedules appointments. Handles interruptions and unclear speech gracefully.

// Voice bot conversation loop
async function voiceConversationLoop(callSession) {
  while (callSession.active) {
    const audio = await callSession.listenForSpeech({
      silenceTimeout: 3000,
      maxDuration: 30000
    });
    const transcript = await speechToText(audio);

    if (!transcript || transcript.confidence < 0.4) {
      await callSession.speak("Sorry, I didn't catch that. Could you repeat?");
      continue;
    }

    const response = await processIntent(transcript.text);
    await callSession.speak(response.text, { voice: "professional-female" });

    if (response.action === "transfer") {
      await callSession.transferTo(response.department);
    }
  }
}

A shopping assistant that asks about preferences, budget, and use case, then recommends products from the catalogue. Uses embeddings to match natural language descriptions to product attributes and handles comparison queries.

// Product recommendation with embeddings
async function recommendProducts(userPreferences) {
  const prefEmbedding = await embed(userPreferences.description);

  const candidates = await vectorStore.similaritySearch(prefEmbedding, {
    filter: { priceRange: userPreferences.budget },
    topK: 10,
  });

  // Re-rank with LLM considering full context
  const ranked = await llm.chat({
    messages: [{
      role: "system",
      content: "Rank these products for the customer. Consider their stated preferences, budget, and use case."
    }, {
      role: "user",
      content: JSON.stringify({ preferences: userPreferences, candidates })
    }]
  });

  return ranked.top3;
}