The recent 10 Days of Voice Agents — VoiceForBharat Edition challenge saw developers tackle real-world problems in India using AI voice technology. Two participants, Naman and Saptak Roy, shared their experiences building voice agents named Shiksha and Roshni, respectively, highlighting the potential of voice interfaces for education and financial inclusion.
Shiksha: A Voice-Based Learning Partner
Naman's project, Shiksha, emerged from a simple idea: to create a voice agent that could help students learn through natural conversation. Over the 10-day challenge, this concept evolved into a comprehensive voice-based learning system. Shiksha is designed to be a voice-first alternative to traditional text-based learning platforms. Students can ask questions, take quizzes, continue their learning sessions with saved profiles, and receive help when they encounter difficulties.
The system goes beyond basic Q&A. Shiksha incorporates features like memory to recall user progress, tools for interactive learning, and a crucial human escalation path for complex queries. For instance, students struggling with mathematics can be seamlessly transferred to a dedicated Maths Specialist agent. This layered approach ensures that students receive support tailored to their needs, whether it’s immediate clarification or specialized tutoring.
The original vision of a conversational learning tool was significantly expanded by the challenge. Naman detailed how Shiksha evolved to include a specialist agent system, allowing for more targeted assistance. The platform aims to make education more accessible and engaging, particularly for those who might find traditional digital interfaces challenging. By allowing students to simply talk to their learning partner, Shiksha removes barriers to entry and fosters a more intuitive learning experience.
Roshni: Bridging the Financial Literacy Gap
Saptak Roy's project, Roshni, addressed a critical issue in India: financial inclusion. Despite rapid digital banking adoption, millions of citizens, especially in tier-2 and tier-3 cities and rural areas, struggle with complex financial terms, interest rates, and government schemes. Text-based interfaces, while common, can be intimidating and inaccessible for many, particularly first-time bank customers or rural entrepreneurs. Saptak recognized that voice is the natural interface for these individuals.
Roshni was built as a real-time, multi-agent financial voice AI. The goal was to provide an ultra-low latency experience, making interactions feel immediate and natural. The system leverages a sophisticated tech stack including Murf Falcon for voice capabilities, LiveKit Agents for multi-agent coordination, Deepgram Nova-3 for advanced speech-to-text, Google Gemini for natural language understanding and generation, and Next.js for the front-end. This combination allows Roshni to handle complex financial queries in multiple languages, making it accessible to a wider population.
The problem statement for Roshni is stark: navigating financial information can be a barrier to economic participation. Saptak highlighted how formal banking terms and eligibility criteria for schemes like PM Mudra or Sukanya Samriddhi Yojana can be daunting. Roshni aims to demystify these complexities. Instead of struggling with online forms or dense documentation, users can simply speak their questions and receive clear, understandable answers in their preferred language. The multi-agent architecture allows Roshni to potentially handle different aspects of financial advice or connect users to specialized financial advisors, mimicking a human interaction flow.
The Technology Behind the Voice Agents
Both Naman and Saptak relied on Murf Falcon, an AI voice platform, as a core component. Murf Falcon likely provided the text-to-speech (TTS) capabilities, enabling the agents to speak in natural-sounding human voices. The challenge's focus on "Bharat" implies a need for regional language support, a feature that advanced TTS platforms can offer.
Beyond TTS, both projects required robust speech-to-text (STT) capabilities to understand user input. Saptak explicitly mentioned Deepgram Nova-3, known for its accuracy and low latency, which is critical for real-time conversational AI. Google Gemini, a powerful large language model, was used by Saptak for understanding context, generating responses, and enabling more sophisticated dialogue management. Naman's Shiksha also implies advanced NLU capabilities to interpret learning-related queries and manage conversational flow.
The "multi-agent" aspect, prominent in Saptak's Roshni and implied in Shiksha's specialist agent system, points to architectures where different AI modules or specialized agents handle distinct tasks. LiveKit Agents, used by Saptak, is a framework designed to facilitate the creation and orchestration of such multi-agent systems. This modular approach allows for scalability and specialization, where one agent might handle STT, another NLU, a third TTS, and yet another might manage business logic or external tool integration. The success of these projects in a 10-day timeframe underscores the increasing maturity and accessibility of AI development tools.
Why Voice Matters for Bharat
The VoiceForBharat initiative rightly identifies voice as a critical technology for India's next wave of digital adoption. In a country with vast linguistic diversity and varying levels of digital literacy, voice interfaces offer a more inclusive and intuitive path to accessing information and services. For education, voice agents like Shiksha can personalize learning and overcome literacy barriers. For financial services, voice agents like Roshni can demystify complex products and empower millions to participate more fully in the digital economy.
The challenge demonstrated that building sophisticated voice agents is now achievable within short hackathon-like timelines, provided developers have access to powerful foundational models and platforms. The focus on real-world problems – education and financial literacy – shows a clear direction for applied AI in emerging markets. The success of Shiksha and Roshni suggests a future where voice AI plays a pivotal role in bridging digital divides and empowering underserved populations across India and beyond.