06/10/2026
🌿 AyurSetu Agentic RAG
I’m sharing my project AyurSetu Agentic RAG — a multilingual AI assistant for Intellectual Property, Ayurveda, Traditional Knowledge, and document-based question answering.
🚀 Key Highlights
🧠 Controlled Agentic RAG
The system analyzes the question, plans retrieval, and decides whether a simple search or deeper multi-query search is required.
📚 ChromaDB Vector Search
Searches both permanent knowledge sources and uploaded documents.
📄 PDF / TXT / DOCX Upload
Documents are extracted, chunked, indexed, and made searchable.
🔤 OCR Capability
Scanned PDFs are processed using Tesseract OCR, PyMuPDF, pytesseract, and Pillow.
⚡ Caching Layer
Uses Exact Cache + Semantic Cache to reduce repeated vector searches and unnecessary AI calls.
🤖 AI Models
Google Gemini as the primary model with Groq fallback.
🌐 Multilingual Support
English, Bengali, and Hindi.
🌿 Ayurveda Centre Search
Integrated location search using SerpAPI.
🔁 CI/CD Pipeline
GitHub Actions → Docker → Railway deployment.
🏗️ Architecture
User
↓
Frontend
HTML / CSS / JavaScript
↓
FastAPI Backend
↓
Agent Router
↓
Agent Service
↓
Query Analysis & Planning
↓
Single Query / Multi Query
↓
ChromaDB Retrieval
├── Permanent Knowledge Base
└── Uploaded Documents
↓
Exact / Semantic Cache
↓
Retrieved Source Context
↓
Gemini
└── Groq Fallback
↓
Multilingual Grounded Answer
↓
Sources + Follow-up Questions
📄 Document Processing Flow
PDF / Scanned PDF
↓
Page-by-Page Processing
↓
Text Extraction / OCR
↓
Chunking
↓
Embeddings
↓
ChromaDB
↓
Searchable Knowledge
🔁 Deployment Flow
Git Push
↓
GitHub Actions
↓
Tests / Build
↓
Docker
↓
Railway
↓
Live Application
🛠️ Tech Stack
🐍 Python
⚡ FastAPI
🧠 Agentic RAG
📚 ChromaDB
🔤 Tesseract OCR
🤖 Gemini
⚙️ Groq
🔎 SerpAPI
🐳 Docker
🔁 GitHub Actions
🚂 Railway
🔗 GitHub
https://github.com/amitava-0304/IP_Shakti_Sahayak
🌐 Live Project
https://web-production-596c7.up.railway.app/
