Last month I happened to see a viral LinkedIn post about SarvamAI (India's AI pioneer) having very few downloads during its initial launch and that no one is asking for a slightly better 24B Indic LLM. So I decided to dive a little deeper to understand what use cases it is trying to solve for. While criticism regarding initial developer downloads or technical claims (e.g., using existing base models like Mistral) is part of the essential scrutiny any tech venture faces, it's crucial not to miss the bigger picture. Here's why I think Sarvam AI (and similar efforts like Krutrim, Hanooman, Soket, Gnani AI, etc.) are incredibly vital for the IndiaAI mission. 1. Beyond English and Translation: It's not just about English proficiency. For India's millions, AI needs to understand and speak in their native languages, including the common "code-mixing" (like Hinglish!). This requires deep linguistic and cultural nuance that global models, despite their power, often lack. This isn't "marginally better"; it's fundamentally more relevant for Bharat. The other day, I was trying to translate a piece of Telugu text into English using ChatGPT, and clearly it misinterpreted many words that changed the whole meaning of that context. Indic models will definitely have an edge in many of these use cases. 2. A National Strategic Imperative: Sarvam AI supports India's AI mission to build our sovereign LLMs for data privacy, national security, and ensuring the nation has indigenous control over this transformative technology. This goes far beyond typical commercial metrics. 3. Enterprise & Government Focus: Their primary focus is on empowering businesses and government services in India. Adoption here isn't about viral downloads, but deep, custom integrations that solve complex, real-world problems for a billion-plus population. Think about the real impact: ℹ️ A farmer getting direct, voice-led agri-scheme info in Punjabi. 🩺 An elderly person understanding health advice in clear Tamil. 📝 A student receiving personalised lessons in Gujarati. 🤖 A multilingual chatbot seamlessly handling "Hinglish" customer queries. 💡 Other examples mentioned in the Sarvam AI cookbook include the regional code helper and doubt solver, travel planner, presentation architect, etc. 4. Building from the Ground Up (and on Shoulders of Giants): Even if models build on existing open-source architectures, the effort to meticulously train and adapt them for India's unique linguistic diversity is a massive R&D undertaking. It's about creating specialised expertise where generic models fall short. India's AI journey is complex, filled with unique challenges and immense potential. We Indians are quick to dismiss our own offerings and tend to hail the West! Let's support and scrutinise these homegrown initiatives with an understanding of the long game – building truly inclusive and impactful AI for everyone. What are your thoughts on this? #SarvamAI #IndiaAI #LLM #TechInIndia
I totally second your thought, sometime the utility and use case driven solutions are very much needed, I was personally experimenting with AI For Bharat models for various use cases and was pleasantly surprised to see they are a lot more context aware specific to Indian context and out perform the models available elsewhere.
💯
Lead Engineer at Wipro||Project: Telstra-Mobile Policy ||Ex-Qualcomm||Mentor at FFE||
1wYes being diverse cultured country we often wanted to use native language when you consider nativity and availability to large-scale India needs robust multilingual Gen AI platform