README.md

metadata

title: RAG for Vrikshayurveda
emoji: 🌿
colorFrom: green
colorTo: green
sdk: gradio
sdk_version: 4.44.1
app_file: app.py
pinned: false
license: mit

🌿 RAG for Vrikshayurveda

Retrieval-Augmented Generation for Surapala's Science of Plant Life (वृक्षायुर्वेद)

A full-stack AI system that diagnoses plant disorders and recommends traditional Ayurvedic treatments based on the ancient Vrikshayurveda text by Surapala. Combines modern RAG architecture with a classical forward + backward chaining inference engine.

---

🏗️ Architecture

User Query
    │
    ▼
┌──────────────────┐
│  Query Expansion  │  ← Groq (Llama 3.3 70B) generates 3 rephrasings
│  (4 variants)     │
└────────┬─────────┘
         │
         ▼
┌──────────────────┐
│  BGE-large-en    │  ← BAAI/bge-large-en-v1.5 (1024-dim embeddings)
│  Encoding        │
└────────┬─────────┘
         │
         ▼
┌──────────────────┐
│  Pinecone Vector │  ← Top-K retrieval + deduplication by parent ID
│  Search          │
└────────┬─────────┘
         │
    ┌────┴────┐
    ▼         ▼
┌────────┐ ┌─────────────────────┐
│ Context│ │  Inference Engine    │
│ Build  │ │  • Symptom Extractor │
│        │ │  • Forward Chaining  │
│        │ │  • Backward Chaining │
│        │ │  • 42 Ayurvedic Rules│
└───┬────┘ └──────────┬──────────┘
    │                 │
    └────────┬────────┘
             ▼
┌──────────────────┐
│  Groq LLM        │  ← Llama 3.3 70B generates final response
│  (RAG + Inference │
│   grounded)       │
└──────────────────┘

---

✨ Features

🩺 Diagnose Tab

• Describe plant symptoms in natural language
• Get Ayurvedic diagnosis with dosha classification (Vata/Pitta/Kapha)
• View the full reasoning chain from the inference engine
• Receive traditional treatment recommendations from Vrikshayurveda
• See retrieved source passages with relevance scores

🔗 Symptom Chain Explorer

• Interactive visual inference engine
• Select symptoms from 27 observable indicators
• Watch forward chaining fire rules in real-time on a canvas
• Step through rules one at a time or auto-run the full chain
• See proof trace with colored condition/conclusion pills
• Stats bar showing fact count, rules fired, and diagnosed disorders

---

🧠 Inference Engine

The inference engine implements 42 Ayurvedic diagnostic rules from the Vrikshayurveda text:

Component	Description
Symptom Extractor	NLP-based extraction of atomic facts from natural language
Forward Chaining	Priority-sorted rule firing to derive disorders and remedies
Backward Chaining	Differential diagnosis across 10 disorder categories
Rule Base	42 YAML-defined rules covering 3 doshas, 10 disorders, 8 causes

Disorder Categories

• Internal: Vata, Pitta, Kapha disorders
• External: Heat/Frost, Stormy Wind, Fire/Lightning/Drought
• Other: Mechanical Wound, Faulty Seed, Insect Infestation, Excessive Watering

---

🚀 Setup

Prerequisites

• Python 3.10+
• API keys for Pinecone, Groq

Installation

git clone https://github.com/Viraj-77/RAG4Vrikshayurveda.git
cd RAG4Vrikshayurveda
pip install -r requirements.txt

Environment Variables

Create a .env file:

GROQ_API_KEY=your_groq_api_key
PINECONE_API_KEY=your_pinecone_api_key
PINECONE_INDEX_NAME=jaim3

Run

python app.py

The app launches at http://127.0.0.1:7860

---

📁 Project Structure

RAG4Vrikshayurveda/
├── app.py                      # Gradio web interface + Symptom Chain Explorer
├── rag_pipeline.py             # Core RAG pipeline (embed → retrieve → generate)
├── inference/
│   ├── engine.py               # Unified inference entry point
│   ├── forward_chain.py        # Forward chaining engine
│   ├── backward_chain.py       # Backward chaining + differential diagnosis
│   └── symptom_extractor.py    # NLP fact extraction from text
├── vrikshayurveda_rules.yaml   # 42 Ayurvedic diagnostic rules
├── db.pdf                      # Vrikshayurveda source text
├── embed_pdf.py                # PDF → Pinecone indexing script
├── augment_and_reindex.py      # Data augmentation + reindexing
├── requirements.txt
└── .env                        # API keys (not committed)

---

🔧 Tech Stack

Component	Technology
Embeddings	BAAI/bge-large-en-v1.5 (1024 dimensions)
Vector DB	Pinecone
LLM	Llama 3.3 70B via Groq
Inference	Custom forward + backward chaining engine
Frontend	Gradio 4.x
Knowledge Base	Vrikshayurveda by Surapala

---

📚 References

• Vrikshayurveda — The Science of Plant Life by Surapala (ancient Indian botanical text)
• Surapala's Vrikshayurveda covers plant diseases, their causes (doshas), and Ayurvedic treatments using natural materials

---

📄 License

This project is licensed under the MIT License.