DEV Community: Pururva Agarwal

GoDavaii Day 10: The Hidden Challenge of Food-Drug Interactions in Multilingual AI

Pururva Agarwal — Tue, 12 May 2026 15:01:37 +0000

40% absorption loss. This isn't a rare anomaly; it's a common, often overlooked reality when certain critical medications, like thyroid pills, are mixed with everyday beverages such as morning tea. This stark fact highlights a massive, unresolved challenge in global health AI: the intricate web of food-drug interactions, especially within diverse cultural and dietary landscapes like India. On Day 10 of building GoDavaii, we're deep-diving into why this problem is technically complex and how we're approaching it for a global, multilingual audience.

Beyond the Pill Bottle: Why Global AI Misses the Kitchen

Most established drug interaction checkers excel at identifying conflicts between pharmaceutical compounds. But their utility often ends at the boundary of the pill bottle. They typically don't account for the vast array of foods, beverages, and traditional remedies that individuals consume daily. In a country like India, where traditional practices ('Desi Ilaaj') often coexist with allopathic medicine, this gap becomes a critical patient safety issue. Imagine the complexity: a common herbal concoction ('kadha') for a cough might interact negatively with a prescribed blood thinner, or a specific fruit could alter the metabolism of a diabetes medication. These aren't edge cases; they are part of daily life for millions. And it's not just food; as Mental Health Awareness Month approaches, we also focus on complex drug-drug interactions, like specific antidepressants with other common medications, which families might overlook.

The Multilingual Maze: Building for India's Health Realities

Addressing food-drug interactions requires more than just a large database; it demands deep cultural and linguistic understanding. Global English-only AI tools fall short here. It's not enough to simply translate "take with food"; the AI Health Companion needs to understand "take with idli-sambar" or "avoid with lassi" in Marathi, Tamil, or Hindi. This involves processing nuanced natural language queries that describe specific local dishes, home remedies, or regional dietary patterns.

At GoDavaii, our AI Health Companion, built to understand 22+ Indian languages, is designed precisely for this. We use advanced models like Gemini 2.5 Flash, not just for raw language translation, but for its multimodal reasoning capabilities to infer cultural dietary patterns and their potential health implications when combined with specific medications. This means building a contextual knowledge graph that bridges medical science with real-world consumption habits, allowing a user in Chennai to ask about their medication and pongal without having to translate their diet into generic English terms.

AI-Verified 'Desi Ilaaj': Navigating Tradition with Technology

Our approach to 'Desi Ilaaj' (AI-verified home remedies) is another layer of this complexity. How do you safely integrate traditional Ayurvedic practices with allopathic medicine, especially concerning potential interactions? This isn't about validating remedies as cures, but about identifying potential conflicts. We undertake a massive effort to digitize, cross-reference, and then AI-validate these remedies against known drug mechanisms, scientific literature, and potential side effects. The challenge lies in the non-standardized nature of traditional knowledge - varying dosages, regional variations, and often undocumented chemical compositions.

Our AI doesn't just list ingredients; it analyzes potential interactions. For instance, if a user queries about a common herbal mix for digestive issues while on an antacid, our system can flag potential efficacy reduction or adverse effects. This is a critical technical challenge: bridging disparate, often qualitative, knowledge systems with quantitative medical data to provide actionable, safety-first insights.

Empowering Families, Augmenting Doctors

GoDavaii is built as a thinking assistant for families. Our Drug Interaction Checker, the AI Health Companion, and features like the Desi Ilaaj verifier are designed to empower individuals with knowledge before their doctor's appointment. We help you surface sharper questions, like "Could my haldi doodh (turmeric milk) affect my prescribed medication?" We do not replace medical professionals; rather, we augment the doctor, helping families catch what a rushed consultation might miss. This is about putting sophisticated, culturally aware health intelligence into the hands of the next billion people coming online in their mother tongues.

What complex interactions, especially involving food or traditional remedies, have you encountered or wondered about? Drop your questions in the comments below, and let's explore this crucial area of health AI together.

The Aunty Test - what Malayalam-speaking patients see when they ask Health AI in their own language

Pururva Agarwal — Tue, 12 May 2026 08:49:10 +0000

Voice query in Malayalam: "thudayil chorachorappu, pus varunnu, enthu cheyyanam?"

Most Health AI built in the last three years is English-first. The translation layer breaks the moment a user types a real query in a language a billion people actually think in.

What a generic Health AI says

You mention itching and pus on thigh. This may be infection. Please consult a doctor. Could you rephrase in English for specific recommendations?

What native multilingual reasoning looks like

Pus varunnathum chorachorappum bacterial infection (likely staph) aakaam. Mupiroin ointment 2% divasam moonnu thavana pukoo. 3 divasathinakam kurayunnillenkil oral antibiotic (Cefadroxil) doctor-um koodi nirdesikkanam.

Why this is the wedge

1.4 billion people don't think in English. Every Health AI built so far has either been English-first (translate-button bolted on) or has had a thin localised veneer over an English model. Neither holds up under a real medical query.

GoDavaii reasons natively in 22 Indian languages including Malayalam, Tamil, Bhojpuri, Marathi, Bengali, Telugu, Kannada, Punjabi, Malayalam, Gujarati, Odia, Urdu and more.

Free on the Play Store: https://play.google.com/store/apps/details?id=com.godavaii.app

Building publicly. Try a query in your home language and tell us where it breaks.

The Aunty Test - what Punjabi-speaking patients see when they ask Health AI in their own language

Pururva Agarwal — Mon, 11 May 2026 10:39:48 +0000

Voice query in Punjabi: "shugar de naal blood pressure di dawai khaani safe hai?"

Most Health AI built in the last three years is English-first. The translation layer breaks the moment a user types a real query in a language a billion people actually think in.

What a generic Health AI says

You seem to be asking about combining diabetes and blood pressure medication. Many combinations are safe but interactions exist. Please rephrase in English for accurate analysis.

What native multilingual reasoning looks like

Metformin te Telmisartan / Amlodipine ekathe khaane safe han - actually diabetic patients vaaste ACE inhibitor (Telmisartan) kidney da bachaa karda hai. Beta blockers (Atenolol) avoid karo agar sugar low ho jaandi hai.

Why this is the wedge

GoDavaii reasons natively in 22 Indian languages including Punjabi, Tamil, Bhojpuri, Marathi, Bengali, Telugu, Kannada, Punjabi, Malayalam, Gujarati, Odia, Urdu and more.

Free on the Play Store: https://play.google.com/store/apps/details?id=com.godavaii.app

Building publicly. Try a query in your home language and tell us where it breaks.

Day 10: What General-Purpose LLMs Miss in Hindi Medical Reasoning for Health AI

Pururva Agarwal — Sun, 10 May 2026 15:01:30 +0000

Audited finding: We tested Hindi medical reasoning on Claude 4. Here is what broke. Not just 'a little off' or 'slightly awkward translation,' but fundamental conceptual errors that would lead to serious health misinformation.

On Day 10 of building GoDavaii, India's Advanced Health AI, the language barrier continues to be the most fascinating and challenging part of our journey. Everyone talks about training general-purpose LLMs on vast datasets, but when you zoom into specific domains like health, and specific languages like Hindi, the cracks become glaring.

The Illusion of 'Multilingual' AI

Most 'multilingual' models are fantastic at translation. Ask them to translate a sentence from English to Hindi, and they'll do a commendable job. But health isn't about translation; it's about context, cultural understanding, and nuanced reasoning. For instance, when we fed Claude 4 (and similar models) complex Hindi medical scenarios - for example, a query about traditional remedies alongside allopathic medication for a specific condition - the results were... concerning. It struggled to cross-verify the safety of 'Desi Ilaaj' (AI-verified home remedies is a core GoDavaii feature) against known drug interactions. It would often miss subtle semantic cues that a native speaker would pick up instantly, especially when describing symptoms in idiomatic Hindi. This isn't a failure of the model's core intelligence, but rather a limitation of its training data's depth in culturally specific medical discourse.

We specifically tested prompts that included questions about common Indian medicines alongside Ayurvedic practices - a reality for millions of Indian families. The models, trained predominantly on English-centric medical texts and broader internet data, often either hallucinated non-existent interactions or, worse, failed to flag genuine ones. They didn't understand the 'why' behind certain traditional practices or the specific terminology used by an aunty in Indore describing her 'kaaichal' (fever) which isn't just 'fever' but often implies a specific kind of malaise.

Building for the Next Billion: Language as a Moat

This isn't just a linguistic challenge; it's an equity one. The 'next billion' users coming online in India will largely do so in their mother tongues. Their health questions won't be posed in English, and their understanding of health isn't solely Western allopathic. For GoDavaii, this isn't a feature; it's the foundation. Our AI Health Companion supports 22+ Indian languages not as a translation layer, but as a deeply embedded understanding engine. We don't just translate 'Drug Interaction Checker' - we ensure the underlying knowledge graph and reasoning engine grasp the nuances of drug names and traditional practices within those language contexts.

World Immunization Week is a powerful reminder of how crucial accessible, accurate health information is. Imagine trying to understand complex vaccination schedules or potential side effects when the explanation is only available in a language you struggle with, or when the AI answering your questions simply doesn't 'get' the cultural context of your concerns. This is the gap we're closing. Our goal is to ensure every medicine you add, every health question you have, is understood and answered with the depth and accuracy it deserves, in the language that matters most to your family.

It's a harder build, no doubt. But the alternative - a health AI that only works for the English-speaking elite - is not the future India deserves. This deep-dive into linguistic and cultural AI challenges continues to shape every architectural decision we make.

What specific language or cultural nuances have you seen general AI tools completely miss in important domains?

The Aunty Test - what Kannada-speaking patients see when they ask Health AI in their own language

Pururva Agarwal — Sun, 10 May 2026 05:28:19 +0000

Voice query in Kannada: "magu ge raatri jwara baruttide, oushadhi kodabeka?"

Most Health AI built in the last three years is English-first. The translation layer breaks the moment a user types a real query in a language a billion people actually think in.

What a generic Health AI says

It seems your child has nighttime fever. Please consult a pediatrician. For more specific guidance, kindly rephrase your concern in English.

What native multilingual reasoning looks like

Magu ge raatri jwara 100.4F daati iddare paracetamol syrup (15mg/kg, 6 ganteg ondu sala) kodbahudu. 3 dina meeri mundhuvariddare CBC + urine test maadisi - UTI ya viral fever differentiate aagutte. Sakkat hydration mukhya.

Why this is the wedge

GoDavaii reasons natively in 22 Indian languages including Kannada, Tamil, Bhojpuri, Marathi, Bengali, Telugu, Kannada, Punjabi, Malayalam, Gujarati, Odia, Urdu and more.

Free on the Play Store: https://play.google.com/store/apps/details?id=com.godavaii.app

Building publicly. Try a query in your home language and tell us where it breaks.

The Aunty Test - what Telugu-speaking patients see when they ask Health AI in their own language

Pururva Agarwal — Sat, 09 May 2026 05:08:42 +0000

Voice query in Telugu: "pregnancy lo paracetamol theesukovacha?"

Most Health AI built in the last three years is English-first. The translation layer breaks the moment a user types a real query in a language a billion people actually think in.

What a generic Health AI says

You appear to be asking about paracetamol in pregnancy. Paracetamol is generally considered safe but please consult your obstetrician. Rephrase in English for more details.

What native multilingual reasoning looks like

Pregnancy lo paracetamol (500mg) generally safe - aithe daily 2g kanna takkuva. First trimester lo NSAIDs (ibuprofen, diclofenac) avoid cheyandi. Fever 100.4F daatithe 24 ganta-lo OB-GYN ni kalavandi.

Why this is the wedge

GoDavaii reasons natively in 22 Indian languages including Telugu, Tamil, Bhojpuri, Marathi, Bengali, Telugu, Kannada, Punjabi, Malayalam, Gujarati, Odia, Urdu and more.

Free on the Play Store: https://play.google.com/store/apps/details?id=com.godavaii.app

Building publicly. Try a query in your home language and tell us where it breaks.

The Aunty Test - what Bengali-speaking patients see when they ask Health AI in their own language

Pururva Agarwal — Fri, 08 May 2026 05:51:43 +0000

Voice query in Bengali: "matha betha ar bomi bomi bhab, ki khabo?"

Most Health AI built in the last three years is English-first. The translation layer breaks the moment a user types a real query in a language a billion people actually think in.

What a generic Health AI says

You mentioned headache and nausea sensation. Common causes include migraine. Please consult a doctor and consider rephrasing in English for clearer medical advice.

What native multilingual reasoning looks like

Matha betha + bomi bomi bhab migraine er classic combination. Andhar ghor, thanda paani-r compress, ar Domperidone (10mg) + Naproxen (250mg) khaowa jete pare. 3 din-er beshi thakle MRI kora uchit.

Why this is the wedge

GoDavaii reasons natively in 22 Indian languages including Bengali, Tamil, Bhojpuri, Marathi, Bengali, Telugu, Kannada, Punjabi, Malayalam, Gujarati, Odia, Urdu and more.

Free on the Play Store: https://play.google.com/store/apps/details?id=com.godavaii.app

Building publicly. Try a query in your home language and tell us where it breaks.

Beyond English: Why AI-Verified Desi Ilaaj is the Unfollowable Moat in Health AI (Day 13/30)

Pururva Agarwal — Thu, 07 May 2026 16:33:12 +0000

Founders often ask me why we didn't just build an English-first Health AI and then add translation layers. The answer lies in phrases like "haldi doodh roz peeyein kya" (should I drink turmeric milk daily?) - a question no English-first model, however advanced, can truly grasp in its cultural and medical context, let alone cross-verify with allopathic medicine. This fundamental misunderstanding isn't a bug; it's a feature of Health AI built for a billion people who don't think in English. And it creates an unfollowable moat.

The Language of Wellness is Not English

Most Health AI doesn't speak the language a billion people think in. We do. While Silicon Valley's Health AI struggles to parse even basic colloquial symptoms like "pet mein gas ho rahi hai" (Hindi for "stomach gas"), the problem deepens when families seek advice on traditional remedies. These queries are deeply embedded in culture and language. An English-first system, even with a sophisticated translation API, fundamentally misses the nuance. It cannot understand the implied context of generations of practice, or the local efficacy debates. The result is a generic, often unhelpful, or even unsafe, response.

For an aunty in Indore, asking about a child's "deham vedanai" (Tamil for "body pain") often comes with an implicit desire to know if a home remedy can alleviate it, before rushing to a doctor. This isn't just about language; it's about a worldview where traditional wisdom and modern medicine coexist. Our AI Health Companion understands this duality, parsing not just the words, but the underlying intent, in 22 Indian languages.

Why Global Health AI Won't Touch Desi Ilaaj

The real moat here is not just technical, but structural. No US or EU-backed Health AI will publicly cross-verify Ayurvedic remedies with allopathic medicine. Their boards, their cap tables, their regulatory environments, and their brand positioning forbid it. Integrating and validating traditional medicine, even with advanced AI, is a line they simply cannot cross. It's a regulatory nightmare and a brand liability in Western markets.

This creates a unique opportunity for GoDavaii. India's cultural depth, where Ayurveda has been practiced for millennia, is a structural advantage that cannot be replicated by companies whose primary market isn't India. Indian families routinely ask the same health question to a doctor and to their grandmother - our AI must respect both sources of wisdom, carefully cross-referencing and providing context, not judgment. This multi-modal, multi-cultural reasoning is where English-first models hit a wall.

The Architectural Bet: Native Multilingual Reasoning

We picked native multilingual reasoning over a translate layer. It cost us a significant portion of our early budget and required far more intricate model design and training data. But it bought us this moat. Instead of translating user queries into English, processing them, and then translating the English output back, we built models that reason directly in Tamil, Hindi, Marathi, Punjabi, and all 22 target languages. This means the AI understands cultural context, colloquialisms, and the subtle interplay between traditional and modern health practices at a native level.

This architectural decision allows our AI Health Companion to provide nuanced insights into Desi Ilaaj, cross-referencing potential drug interactions with allopathic medicines. For example, if a user queries a specific herb and also lists prescription drugs, GoDavaii's interaction checker will flag potential conflicts, something an English-first tool would entirely miss or misinterpret. It's about providing a fresh pass over your medicine cabinet that understands the full spectrum of a family's health context.

Day 13/30 building GoDavaii, we're doubling down on this approach. The future of Health AI for the Next Billion isn't about adapting English models; it's about building from first principles in the languages and cultural contexts where they live and think.

What structural moats are you seeing in your industry that Big Tech cannot or will not touch?

Try GoDavaii and explore our unique multilingual capabilities at godavaii.com. Your feedback is invaluable as we continue this journey.

The Aunty Test - what Marathi-speaking patients see when they ask Health AI in their own language

Pururva Agarwal — Thu, 07 May 2026 05:04:54 +0000

Voice query in Marathi: "sugar chi gholi ghetli tari BP high yetoy, kay karu?"

Most Health AI built in the last three years is English-first. The translation layer breaks the moment a user types a real query in a language a billion people actually think in.

What a generic Health AI says

It looks like your BP is high despite diabetes medication. Diabetes and hypertension are different conditions. Please rephrase in English so I can provide accurate guidance.

What native multilingual reasoning looks like

Diabetes chi golya BP la control karat nahit - donhi vegli aushad lagtat. Tumcha BP reading kay aahe? 140/90 paeksha jaast astil tar Telmisartan ki Amlodipine baddal doctor-shi boluya. Diabetic asal tar ACE inhibitor preferred.

Why this is the wedge

GoDavaii reasons natively in 22 Indian languages including Marathi, Tamil, Bhojpuri, Marathi, Bengali, Telugu, Kannada, Punjabi, Malayalam, Gujarati, Odia, Urdu and more.

Free on the Play Store: https://play.google.com/store/apps/details?id=com.godavaii.app

Building publicly. Try a query in your home language and tell us where it breaks.

The Aunty Test - what Bhojpuri-speaking patients see when they ask Health AI in their own language

Pururva Agarwal — Wed, 06 May 2026 05:02:33 +0000

Voice query in Bhojpuri: "khansi do hafta se hai, ratiya mein zaada hot ba."

Most Health AI built in the last three years is English-first. The translation layer breaks the moment a user types a real query in a language a billion people actually think in.

What a generic Health AI says

It appears you have a persistent cough for two weeks worsening at night. Please see a doctor. Could you rephrase your symptoms in standard Hindi or English for more help?

What native multilingual reasoning looks like

Do hafta se khansi aur ratiya mein badhe wala matlab acid reflux ya post-nasal drip ho sakela. TB ke ruling out karaeke khatir chest X-ray + sputum test karaval ja. Honey + ginger short term raahat di.

Why this is the wedge

GoDavaii reasons natively in 22 Indian languages including Bhojpuri, Tamil, Bhojpuri, Marathi, Bengali, Telugu, Kannada, Punjabi, Malayalam, Gujarati, Odia, Urdu and more.

Free on the Play Store: https://play.google.com/store/apps/details?id=com.godavaii.app

Building publicly. Try a query in your home language and tell us where it breaks.

The Aunty Test - what Tamil-speaking patients see when they ask Health AI in their own language

Pururva Agarwal — Tue, 05 May 2026 06:30:04 +0000

Voice query in Tamil: "kaaichal moonu naal aagudhu, paracetamol thaan podhuma?"

Most Health AI built in the last three years is English-first. The translation layer breaks the moment a user types a real query in a language a billion people actually think in.

What a generic Health AI says

I notice you mentioned fever for three days. Paracetamol may help. For accurate Tamil medical guidance, please consult a Tamil-speaking physician or rephrase your query in English.

What native multilingual reasoning looks like

Moonu naal mela kaaichal irundha paracetamol madhum podhadhu. Dengue, typhoid, malaria check pannanum. CBC + dengue NS1 test edukkanum. Indha 24 mani neram-la doctor-a paarunga.

Why this is the wedge

GoDavaii reasons natively in 22 Indian languages including Tamil, Tamil, Bhojpuri, Marathi, Bengali, Telugu, Kannada, Punjabi, Malayalam, Gujarati, Odia, Urdu and more.

Free on the Play Store: https://play.google.com/store/apps/details?id=com.godavaii.app

Building publicly. Try a query in your home language and tell us where it breaks.

The Aunty Test - what Kannada-speaking patients see when they ask Health AI in their own language

Pururva Agarwal — Mon, 04 May 2026 14:07:29 +0000

Voice query in Kannada: "magu ge raatri jwara baruttide, oushadhi kodabeka?"

Most Health AI built in the last three years is English-first. The translation layer breaks the moment a user types a real query in a language a billion people actually think in.

What a generic Health AI says

It seems your child has nighttime fever. Please consult a pediatrician. For more specific guidance, kindly rephrase your concern in English.

What native multilingual reasoning looks like

Magu ge raatri jwara 100.4F daati iddare paracetamol syrup (15mg/kg, 6 ganteg ondu sala) kodbahudu. 3 dina meeri mundhuvariddare CBC + urine test maadisi - UTI ya viral fever differentiate aagutte. Sakkat hydration mukhya.

Why this is the wedge

GoDavaii reasons natively in 22 Indian languages including Kannada, Tamil, Bhojpuri, Marathi, Bengali, Telugu, Kannada, Punjabi, Malayalam, Gujarati, Odia, Urdu and more.

Free on the Play Store: https://play.google.com/store/apps/details?id=com.godavaii.app

Building publicly. Try a query in your home language and tell us where it breaks.