Three months. That's how long I've been heads-down on just one core problem at GoDavaii: modeling drug interactions. In medical school, this might be a single module, maybe a week's worth of content. But building a health AI for the sheer complexity of Indian family health, for 22+ languages, and for genuine cross-verification of allopathic and traditional remedies? That's a different beast entirely.
I'm Pururva Agarwal, founder of GoDavaii, and we're building India's Advanced Health AI. Today is Day 17 of our public sprint, and while we're still refining, the depth of this particular challenge often surprises people outside of health tech. It's not just about a database; it's about context, culture, and cognition.
The Textbook vs. The Real World: Why a Week Isn't Enough
Standard drug interaction checkers, even good ones like Epocrates, often rely on structured, well-documented clinical data. This is foundational. However, the moment you step outside the most common 500 drugs, or the moment a patient isn't in a perfectly controlled clinical trial, the landscape shifts. For instance, a persistent cough - a common symptom that CARE Hospitals doctors recently highlighted as a potential indicator of undiagnosed asthma - often leads to multiple consultations and, potentially, multiple prescriptions from different doctors. This layered prescribing dramatically increases the risk of unforeseen interactions. Our interaction checker needs to account for every medicine you add, not just a static, pre-filtered list.
My focus for these past three months has been on architecting a system that doesn't just match drug names but understands the active pharmaceutical ingredients, their various forms (generics, brands, fixed-dose combinations common in India), and crucially, how they metabolize together. This isn't a simple lookup; it's a dynamic graph problem. We're building a knowledge graph where nodes aren't just medicines, but also conditions, symptoms, and potential interactions, all with varying degrees of certainty and clinical significance.
Beyond the Pill: Integrating Desi Ilaaj and Multilingual Complexity
Here's where GoDavaii truly differentiates. While global competitors are primarily English-only and focused solely on allopathic medicine, India's health landscape is far richer. Consider the reality: someone takes an allopathic blood pressure medication and a traditional Ayurvedic decoction for general wellness. How do these two interact? Conventional systems don't even see the Desi Ilaaj (AI-verified home remedies) part of the equation.
We've spent months building out datasets and training models to understand these cross-system interactions. It's not about replacing traditional knowledge, but about AI-verifying it, flagging potential conflicts, and empowering families to ask better-targeted questions of their doctors. This is genuinely unique globally, and it's a significant part of why we're building for the 'next billion' - people coming online in their mother tongue with health questions English AI cannot answer.
Then there's the language. Our AI Health Chat operates in 22+ Indian languages. The challenge: identifying a specific Ayurvedic herb, or a complex symptom description like 'tabiyat theek nahi' in Tamil, and then mapping it reliably to a medical concept that can be checked against a drug interaction graph. This requires deep linguistic understanding, not just translation. It's a fundamental challenge for health AI, and one we've invested heavily in, including use models like Gemini 2.5 Flash for the nuanced interpretations needed.
Building the Graph: A Thinking Tool, Not a Diagnosis
This intricate modeling is precisely why GoDavaii isn't a substitute for your doctor. Instead, we're building a powerful question-builder for families. It's an extra check before your next appointment, helping you surface questions, understand potential risks, and engage more deeply with your healthcare provider. We're not making diagnoses or guaranteeing cures; we're augmenting the family's ability to navigate a complex health system.
Our approach to building this interaction graph involves constant validation and iterative learning. We're not just ingesting data; we're building a system that can reason over it, identify patterns, and highlight ambiguities. This foundational work is what allows us to offer features like our Pregnancy medicine safety checker or Lab Report AI explanation - each relying on a robust understanding of how different medical entities interrelate.
What's Next: Day 17 and the Road Ahead
Reaching the Top 14 Global Finalist in Startup Flight Vietnam 2025 was a tremendous validation for GoDavaii, and it highlighted that the global tech community recognizes the scale and uniqueness of this problem. But the real work is in the trenches, continuing to refine these core models.
Day 17 of 30 is a checkpoint, not a finish line. The goal isn't just a functional interaction checker; it's a trustworthy, intelligent assistant that speaks to India's unique health realities. If you're building in health tech, or even just curious about the complexities, what's been your biggest challenge in translating 'textbook' knowledge into real-world applications? Drop your questions in the comments below, I'd love to hear your thoughts.
Top comments (0)