We're on Day 5 of our public sprint building GoDavaii, and we've just crossed 379 users. It's a small but meaningful start on our way to supporting 100,000 families. But for me, Pururva Agarwal, the 27-year-old founder, the real milestone isn't just the user count - it's every single time our AI Health Chat correctly understands a symptom described in, say, Marathi, exactly the way my grandmother would phrase it. This isn't just about translating English into 22 Indian languages; it's about understanding 22 distinct ways a family describes 'not feeling well.' And honestly, that's where most global health apps, built for English-first markets, completely miss the mark.
The Nuance of "Kaaichal" - Beyond Direct Translation
When someone says "I have a fever" in English, it's pretty straightforward. But in India, the same underlying condition might be described as 'kaaichal' in Tamil, 'tap' in Hindi, or 'jwar' in Marathi. The challenge isn't just mapping these words; it's about the context and implied severity that often comes with them. My grandmother, for instance, might say, "Thoda tap aala aahe, potat dukhta aahe" (I have a bit of a fever, and my stomach hurts). An English-centric AI, even with a translation layer, might struggle to piece together the nuance of "thoda tap" or the casual way serious symptoms are sometimes conveyed.
Global competitors like Epocrates or drugs.com are incredible resources, but they're fundamentally built on English medical terminologies and Western healthcare contexts. They don't account for the reality of multi-generational Indian homes where medical conversations happen in regional languages, often mixing traditional beliefs with modern medicine. Our approach with GoDavaii's AI Health Chat, use models like Gemini 2.5 Flash, isn't just about language processing. It's about cultural processing - understanding that a mother in Punjab describing her child's cough isn't using clinical terms, but a rich, localized vocabulary that needs intelligent interpretation.
The Data Problem: Training for True Linguistic Diversity
Building an AI that genuinely understands 22+ Indian languages, beyond a mere Google Translate wrapper, is a monumental data challenge. Where do you find high-quality, medically-relevant conversational datasets in these languages? The answer: you build them. We've had to work with linguists and medical experts across different regions to curate and annotate conversations, focusing on how symptoms, medical histories, and traditional remedies (our AI-verified Desi Ilaaj feature being a prime example) are actually discussed. This isn't just about feeding in dictionaries; it's about teaching the AI to infer, to ask clarifying questions in the right language, and to recognize regional colloquialisms that signify specific health concerns.
For a technical audience, imagine the complexity of training a robust NLP pipeline not just for different languages, but for different styles of medical communication within those languages. We're talking about everything from the subtle differences in expressing abdominal discomfort in Bengali versus Gujarati, to understanding the context of fasting traditions (like during Karva Chauth or Navratri) that might influence medication schedules or dietary advice.
GoDavaii's Approach: An Augmented Thinking Tool
GoDavaii is built to be a thinking assistant for families. It helps them surface questions to ask their doctor, providing a second pair of eyes before their next appointment and catching what a rushed consultation might have missed. It's not a substitute for your doctor, and it certainly doesn't make diagnoses. My goal, inspired by my own grandmother's daily medication challenges, is to empower families to be better advocates for their own health, regardless of the language they speak at home.
Our Drug Interaction Checker, for example, is not just running an English database through a translator. It's designed to understand medicine names as they're commonly used in India, and it's backed by our language-aware AI. The same goes for our Pregnancy medicine safety checker or the Lab Report AI explanation - all designed to bridge the language gap.
We're just on Day 5, but every day we get closer to understanding the true linguistic mosaic of Indian healthcare. It's a difficult path, but it's the only way to build something that genuinely serves families like mine. We're not just adding features; we're rebuilding the foundation of health AI for an entirely different reality.
What's the most challenging localization problem you've tackled in your own projects? I'd love to hear your insights in the comments. You can explore GoDavaii's approach at godavaii.com.
Top comments (0)