{
  "id": 1030697,
  "title": "Sydney: Building a Hinglish AI/ML Mentor That Actually Talks — 10 Days of Voice Agents",
  "url": "https://urgent.news/2026/08/15/sydney-building-a-hinglish-ai-ml-mentor-that-actually-talks-10-days",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-15T11:50:50.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/suyash_sahu_71d79c724bbdf/sydney-building-a-hinglish-aiml-mentor-that-actually-talks-10-days-of-voice-agents-1gdb"
  },
  "original_language": "en",
  "account": "During a 10-day challenge called Voice for Bharat Challenge 2026, one participant built an AI/ML mentor named Sydney. Sydney is an AI companion designed to teach concepts like RAG, backpropagation, embeddings, and agent architectures through voice conversations. The mentor has several features that allow learners to interact with it naturally in Hinglish, a mix of Hindi and English.\n\nSydney can remember returning learners, fetch practice exercises tailored to their level, make outbound calls for scheduled daily practice sessions, escalate to a human when a learner is genuinely struggling, track her own success rate on a live dashboard, and hand off to specialist agents when a question needs deeper expertise. This post explains what Sydney does, the key features, the bugs encountered during development, and provides enough practical details for others to build their own voice agent.\n\nThe primary challenges for AI/ML learning material are that most content is written in technical English, which is difficult for many learners in India who think and explain concepts in Hinglish. Text-based tutorials often let users feel they understand something without actually explaining it back. Voice conversations, however, make it clear when there is a gap in understanding, as learners must explain concepts back in real-time.\n\nSydney's primary tools include speech-to-text (STT), a language model (LLM), and text-to-speech (TTS). Deepgram's nova-3 model handles STT, supporting both Hindi and English. Google's gemini-3.5-flash-lite LLM powers the conversation, with Murf Falcon TTS converting text to speech using an Anisha voice in Conversation style. LiveKit Agents manages the WebRTC session, turn detection, and voice activity detection.\n\nThe most important features of Sydney include a live wave-visualizer frontend that shows connection status, persistent memory of returning learners with name, language preference, and topics covered, a practice exercise tool that fetches level-appropriate questions, outbound calling for scheduled practice sessions, human escalation when a learner is distressed, and specialist handoff to deeper expertise when needed.",
  "summary": "One day seating in the on the TL;DR Over 10 days, as part of Murf AI's 10 Days of Voice Agents — Voice for Bharat Challenge 2026 , I built Sydney — a Hinglish-speaking AI/ML learning companion that teaches concepts like RAG, backpropagation, embeddings, and agent architectures entirely through voice conversation. She remembers returning learners, fetches practice exercises, makes outbound calls,…",
  "key_points": [
    "Sydney AI mentor teaches RAG, backpropagation, embeddings, agent architectures",
    "Sydney conversational AI in Hinglish, Hindi-English mix",
    "Sydney features include memory, practice exercises, human escalation"
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "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."
}