Care AI: India-aware Medical Chatbot using LangChain & RAG

• 

  LocalCare AI – An India-aware medical chatbot powered by LangChain & RAG for safe, localized healthcare guidance.

Inspiration: Healthcare misinformation is a major challenge in India, especially in rural and semi-urban areas where professional advice isn’t always accessible. Many rely on random web searches, which often causes panic or delays in seeking care.

What it does: LocalCare AI is an India-aware medical chatbot built using LangChain + Retrieval-Augmented Generation (RAG). It: Provides evidence-backed answers to common health queries. Offers triage guidance (self-care / see doctor / emergency). Connects patients with local healthcare resources (Ayushman Bharat schemes, vaccination info, nearby clinics). Uses guardrails to block unsafe or diagnostic responses, escalating emergencies to human care.

How we built it: Collected public healthcare documents (WHO, MOHFW, PubMed abstracts). Preprocessed, chunked, and embedded texts into a vector database (Qdrant/Weaviate). Built a LangChain RAG pipeline to retrieve top results and generate answers with citations. Added safety filters to detect red-flag queries and prevent unsafe instructions. Designed a simple chat UI frontend with source highlighting.

Challenges i ran into: Balancing accuracy with speed. Ensuring safety (no hallucinated medical prescriptions). Collecting India-specific datasets and guidelines in structured form.

Accomplishments we’re proud of: Built a functional medical chatbot that grounds answers in real documents. Added triage + local healthcare integration, making it relevant for India. Created a safe system with guardrails and escalation pathways.

What we learned: How LangChain RAG pipelines reduce hallucinations. Importance of ethical design in healthcare AI. Building end-to-end systems in a short time using modular tools.

What’s next: Expand to regional languages with voice support. Integrate with ABHA IDs under India’s ABDM. Partner with NGOs / state health bodies for real-world pilot testing.

Built With

• deeplearning
• hugging-face-embeddings
• langchain
• openai-api
• python
• qdrant
• rag
• streamlit/react
• vercel