{
  "id": 6219028,
  "title": "What we learnt building a voice agent on two frontier Indian AI models",
  "url": "https://urgent.news/2026/09/08/what-we-learnt-building-a-voice-agent-on-two-frontier-indian-ai-models",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-08T02:30:10.000Z",
  "source": {
    "name": "The Ken",
    "slug": "the-ken",
    "url": "https://the-ken.com/story/what-we-learnt-building-a-voice-agent-on-two-frontier-indian-ai-models/"
  },
  "original_language": "en",
  "account": "On a weekend, two programmers built a voice-transcription AI agent using India's state-funded large language model BharatGen and compared it with US-based Eleven Labs' Scribe v2. BharatGen was slow, glitchy, and inaccurate, while Sarvam and Eleven Labs had faster processes and higher accuracy. Maharashtra signed agreements with both companies for government use, while Karnataka explored Eleven Labs' voice AI. BharatGen's AI assistant MahaGPT remains in pilot stage.",
  "summary": "The post What we learnt building a voice agent on two frontier Indian AI models appeared first on The Ken .",
  "key_points": [
    "Two programmers built voice-transcription AI agent using BharatGen and Eleven Labs.",
    "BharatGen was slow, glitchy, and inaccurate compared to Eleven Labs and Sarvam.",
    "Maharashtra signed agreements with both companies for government use."
  ],
  "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."
}