Sarvagya AI 🧠

Sarvagya AI is a localized, enterprise-grade Retrieval-Augmented Generation (RAG) chatbot system designed to securely process, query, and retrieve institutional knowledge for Dr. Shyama Prasad Mukherjee University (DSPMU). Built on Spring Web MVC and modern Java concurrency, it leverages local LLMs to ensure strict data privacy while maintaining high-throughput inference for the System.

🚀 Key Features

• Automated ETL Knowledge Pipeline: A robust Extract, Transform, Load workflow that ingests unstructured data (documents, notices), applies semantic chunking, generates embeddings, and indexes them for rapid retrieval.
• Local LLM Inference: Completely private, local text generation and tool-calling powered by Ollama, eliminating reliance on external APIs.
• High-Performance Vector Search: Utilizes PostgreSQL with pgvector and HNSW indexing for low-latency, high-dimensional similarity searches.
• Virtual Thread Concurrency: Utilizes Java 21 Virtual Threads within Spring Web MVC to efficiently handle blocking LLM and database I/O, preventing thread starvation under high concurrent load.
• Agentic Workflows: Integrates the Model Context Protocol (MCP) to allow the AI to interact with external tools and APIs dynamically.
• Containerized Deployment: Fully dockerized ecosystem for seamless orchestration of the database, backend, and LLM inference engine.

🛠️ Tech Stack

• Backend: Java 21, Spring Boot, Spring Web MVC, Spring AI
• Database: PostgreSQL, pgvector extension
• AI / LLM: Ollama (Llama 3 / Mistral), Model Context Protocol (MCP)
• DevOps & Deployment: Docker, Docker Compose
• Build Tool: Maven

🔄 Architecture & ETL Flow

1. Extract: Raw institutional documents and text are ingested into the Spring Boot backend.
2. Transform: The text is cleaned and passed through a semantic text splitter. Spring AI then interfaces with a local embedding model to convert these chunks into dense vector representations.
3. Load: The vectors and their associated metadata are stored in PostgreSQL (pgvector), utilizing an HNSW index to optimize for cosine similarity.
4. Retrieve & Generate (RAG): When a user submits a query, it is embedded and matched against the vector database. The most relevant chunks are retrieved and injected into the system prompt for the local LLM (via Ollama) to generate a context-aware, hallucination-free response.

⚙️ Local Setup & Installation

Prerequisites

• Docker & Docker Compose
• Java 21+
• Maven

Running the System

1. Clone the repository:

   git clone [https://github.com/your-username/sarvagya-ai.git](https://github.com/your-username/sarvagya-ai.git)
   cd sarvagya-ai