Swasthya AI

🚀 Inspiration

Healthcare knowledge shouldn’t depend on one’s language or literacy level. In rural communities, people often wait until an illness becomes critical because preventive guidance and trustworthy information are unavailable in their dialects. Many can’t read long text or navigate English‑based applications. Swasthya AI was born from the idea that a warm, familiar voice in your own language can make healthcare more accessible, empathetic, and timely.

💡 What It Does

Swasthya AI is a voice‑based multilingual health assistant that:

Listens to questions spoken or typed in Hindi, Hinglish, or local dialects. Provides safe, verified home remedies and preventive‑care advice instantly. Gently alerts users when symptoms require medical attention. Speaks back in natural, clear Hindi using Text‑to‑Speech technology. In short, it transforms AI and language models into a caring, conversational health companion for underserved communities.

🧠 How We Built It

Backend: Python + FastAPI for quick, async routing of queries. Language Understanding: LLaMA model fine‑tuned for Hindi/Hinglish intent detection. Knowledge Source: Cerebras AI Platform as the verified‑content backbone—pulling accurate, contextual health information into a local JSON store. Orchestration: MCP Gateway connects the model, the knowledge base, and the messaging layer seamlessly. Voice Output: Google Cloud Text‑to‑Speech converts responses into regional‑accented Hindi audio. User Interface: WhatsApp‑style chat or web endpoint that works even on low bandwidth. Everything runs on lightweight servers, making it deployable for outreach programs and low‑resource devices.

⚙️ Challenges We Ran Into

Translating mixed‑language and phonetically spelled queries into medically relevant intents. Balancing helpful home remedies with strong safety guardrails (to avoid misuse or overconfidence). Compressing multiple cutting‑edge APIs—LLMs, Cerebras, MCP routing, and TTS—into a single weekend prototype. Testing multilingual voice synthesis quality across different dialect pronunciations.

🏆 Accomplishments That We’re Proud Of

Built a fully functioning end‑to‑end prototype as a solo developer within hackathon time. Achieved natural dialogue understanding for real Hinglish phrases. Integrated the Cerebras knowledge base successfully for fast, trusted responses. Delivered output in a calm, reassuring Hindi voice—making technology genuinely feel human.

📚 What We Learned

How to combine multiple AI services safely through a structured gateway (MCP). How large‑language models like LLaMA can be fine‑tuned for specific cultural and linguistic contexts. The importance of human‑centered design—users value friendliness and clarity more than technical jargon. Real insight: Responsible AI isn’t about perfection—it’s about empathy and transparency.

🔮 What’s Next for Swasthya AI

Add more regional languages—Bengali, Tamil, Marathi, Bhojpuri. Build a mobile PWA with offline caching for zero‑data zones. Partner with government health missions and NGOs to deliver preventive‑care campaigns. Integrate basic vital‑symptom tracking and connect with telemedicine APIs for escalation. Continue crowdsourcing local myth‑busting content to keep the knowledge base fresh and relevant.

Built With

• cerebras
• llama
• next.js
• react-native
• typescript