csp.
HomeAboutWorkResearchBlogContact
Back to Projects

AyurMind

AyurMind is a specialized RAG system designed to retrieve and generate insights from Ayurvedic medical texts. It combines dense retrieval with domain-specific embedding models to provide accurate, context-aware answers about Ayurvedic medicine, treatments, and formulations.

PythonLangChainChromaDBOpenAIFastAPIStreamlitDocker

Problem

Ayurvedic knowledge is scattered across ancient texts in Sanskrit and regional languages, making it inaccessible to modern practitioners and researchers.

Solution

Built a domain-specific RAG pipeline that chunks, embeds, and retrieves from digitized Ayurvedic texts, then uses GPT-4 to generate contextual responses.

Architecture

  1. Text extraction: OCR pipeline for digitized manuscripts
  2. Chunking: Semantic chunking with overlap for context preservation
  3. Embeddings: Fine-tuned Sentence-BERT on Ayurvedic corpus
  4. Vector store: ChromaDB for efficient similarity search
  5. Retrieval: Hybrid search (dense + BM25) with re-ranking
  6. Generation: GPT-4 with domain-specific prompts

Results

90% retrieval accuracy on domain-specific queries, deployed as a web app with Streamlit, used by 50+ Ayurvedic practitioners.

FAQs

How accurate are the responses?

The system achieves 90% retrieval accuracy. All responses include source citations from the original texts for verification.