{
  "id": 978488,
  "title": "Building Samar: My 10-Day Voice AI Agent Journey with Murf Falcon",
  "url": "https://urgent.news/2026/08/15/building-samar-my-10-day-voice-ai-agent-journey-with-murf-falcon",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-15T06:22:37.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/raghul_p23itr122_8708ad5/building-samar-my-10-day-voice-ai-agent-journey-with-murf-falcon-41no"
  },
  "original_language": "en",
  "account": "In a span of ten days, I constructed Samar, a multilingual AI voice agent tailored for a Bharat Digital Bank use case as part of the 10 Days of Voice Agents – VoiceForBharat Edition challenge. Initially conceived as a basic voice assistant, Samar evolved into a sophisticated Voice AI system with the ability to recognize users, leverage real-time tools, initiate outbound calls, delegate sensitive matters to human support, evaluate calls, and forward specialized conversations to another AI agent.\n\nBanks often pose challenges for customers due to the need for quick information or assistance without navigating through multiple screens and menus. Samar was designed to be a voice-first banking assistant that offers natural conversations while ensuring security and identifying when to involve a human. This is where Samar comes into play.\n\nSamar is a multilingual banking voice agent intended to assist users with general banking inquiries. Its capabilities include:\n\n- Addressing common banking questions.\n- Offering financial information.\n- Recognizing returning users with consent.\n- Acquiring real-time data using tools.\n- Locating nearby branches.\n- Providing exchange-rate details.\n- Initiating reminder calls via outbound communication.\n- Escalating delicate issues to human support.\n- Monitoring call analytics.\n- Assigning specialized conversations to a dedicated agent.\n\nThe system's voice experience is facilitated by Murf Falcon, the quickest Text-to-Speech (TTS) API involved in the challenge.\n\nThe interaction between the user and Samar unfolds as follows: User Speech → Speech-to-Text → LLM / Agent Logic → Memory or Tool Calling → Text-to-Speech → User hears the response. The system employs real-time voice communication via LiveKit, uses an LLM for reasoning and conversation, conducts speech recognition to understand the user, and generates natural voice through Murf Falcon.\n\nKey features of Samar include:\n\n1. Voice AI with Guardrails: Samar adheres to safety rules, refraining from requesting sensitive information like PINs, OTPs, CVVs, and full account credentials. It acknowledges its limitations when faced with sensitive requests.\n\n2. Multilingual Conversations: Samar supports multilingual interactions, observing native-script requirements. For instance, Hindi uses Devanagari script, while English employs the Latin script, enhancing the naturalness and accuracy of multilingual responses.\n\n3. Persistent Memory: Samar can remember returning users upon explicit consent. The system retains pertinent profile information and retrieves it during subsequent conversations, enabling a more personalized user experience.\n\n4. Real-Time Tools: Samar is equipped with tools that enable it to access pertinent information independently of the LLM. These tools include:\n- Exchange-rate lookup\n- Nearby branch identification\n- Banking-related data\n\nThe agent determines when a tool is necessary and presents the result to the user in a natural manner.\n\n5. Outbound Calls: Samar possesses the capability to initiate outbound calls rather than merely waiting for users to reach out. One application is a banking reminder where Samar proactively contacts the user with essential updates.\n\n6. Human Escalation: An effective AI agent should recognize when it cannot safely manage a situation. For example, if a user reports potential fraud, Samar can seek permission to generate a support request and forward the issue to human assistance.\n\n7. Call Analytics: A Call Analytics Dashboard has been developed to gain insights into Samar's performance. The dashboard can offer information such as:\n- Total calls\n- Completed calls\n- Human escalations\n- Average call duration\n- Tool usage\n- Call trends\n- Language distribution\n- Recent escalations\n\nThis enables a better understanding of user interactions and facilitates improvements to the agent.\n\n8. Specialist Agent Handoff: Samar does not claim expertise in all areas. When confronted with specialized inquiries, the main agent can transfer the conversation to a specialized agent. For instance:\n- User → Samar → Loan Specialist\n\nThe specialized agent can then continue the discussion with more targeted knowledge.",
  "summary": "Building Samar: My 10-Day Voice AI Agent Journey with Murf Falcon Over the past 10 days, I built Samar , a multilingual AI voice agent for a Bharat Digital Bank use case as part of the 10 Days of Voice Agents – VoiceForBharat Edition challenge. The project started as a simple voice assistant and gradually evolved into a more complete Voice AI system capable of remembering users, using real-time…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "Dev.to",
        "title": "Building Vyapar AI: My 10-Day Journey Building an Indian Voice Agent with Murf Falcon",
        "url": "https://urgent.news/2026/08/15/building-vyapar-ai-my-10-day-journey-building-an-indian-voice-agent",
        "published": "2026-08-15T05:08:12.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}