🚀 SRM Insiders Agentic RAG System

An intelligent, stateful Retrieval-Augmented Generation (RAG) assistant for the SRM Insiders Club.

The SRM Insiders Agent is an advanced AI assistant designed to provide club members with precise, factual answers regarding club policies, roles, and deadlines. Built with LangGraph, it queries a centralized database of official club manuals to guarantee accurate and hallucination-free responses.

✨ Key Features

• Strict Factual Accuracy: Utilizes local vector search (ChromaDB) to retrieve exact context from club manuals before answering.
• Dual Deployment Architecture:
  • ☁️ Cloud Mode (Groq): Lightning-fast inference via llama-3.1-8b-instant, optimized for scalable deployment on Streamlit Community Cloud.
  • 💻 Local Mode (Ollama): 100% offline local inference for maximum privacy, utilizing llama3 with automated fallbacks to qwen2.5 and phi3.
• Stateful Memory: Maintains conversational context using LangGraph's state machine to handle complex, multi-turn follow-up questions.
• Modern UI: Clean, interactive chat interface built on Streamlit.

---

📊 System Architecture & Workflows

High-Level System Architecture

The system is designed to dynamically route between cloud and local LLMs while relying on a centralized local vector database for factual grounding.

flowchart TD
    User([Club Member]) -->|Ask Question| Streamlit[Streamlit Chat UI]
    Streamlit -->|Updates State| LangGraph[LangGraph State Machine]
    
    subgraph "Execution Layer"
        LangGraph -->|USE_LOCAL_LLM=False| Groq[Groq API: Llama-3.1]
        LangGraph -->|USE_LOCAL_LLM=True| Local[Ollama: Llama3 / Qwen2.5 / Phi3]
    end
    
    subgraph "Knowledge Base Retrieval"
        Groq -.->|Calls Search Tool| RAG[search_club_policies]
        Local -.->|Calls Search Tool| RAG
        RAG -->|Semantic Query| Chroma[(ChromaDB)]
        Chroma -->|Returns Top-K Docs| RAG
    end
    
    RAG -.->|Context Injected| LangGraph
    LangGraph -->|Generates Answer| Streamlit

Loading

LangGraph State Machine Loop

The core intelligence of the agent operates on a cyclical node-graph. It continuously evaluates whether it has enough information to answer the user, or if it needs to trigger a database search.

stateDiagram-v2
    [*] --> agent_brain
    
    agent_brain --> _should_continue : Evaluate Next Step
    
    state _should_continue {
        direction LR
        Need_Information --> execute_tools
        Has_Answer --> [*]
    }
    
    execute_tools --> agent_brain : Return Document Context

Loading

---

🛠️ System Requirements

• Python 3.13 or higher
• A free Groq API Key (for cloud inference). Obtainable at the Groq Console.

---

🚀 Installation & Setup

1. Clone the repository:

git clone https://github.com/shreyascode11/Insider-Agent.git
cd Insider-Agent

2. Install dependencies: We recommend using uv for significantly faster dependency resolution.

uv sync 
# Alternatively: pip install -r requirements.txt

3. Ingest your knowledge base: Place your club's official documentation (.txt or .md files) into the data/raw_docs/ directory. Then, generate the vector embeddings:

uv run python ingest.py

This will process the documents and build the persistent ChromaDB vector store locally.

---

🎮 Execution Modes

Mode 1: Cloud Deployment (Recommended for Production)

This mode leverages Groq's APIs for near-instantaneous response generation, making it the ideal choice for hosting a public URL for club members.

1. Create a .env file in the root directory and add your API key:

GROQ_API_KEY=gsk_your_api_key_goes_here

2. Launch the Streamlit server:

uv run streamlit run app.py

Mode 2: Local Offline Inference (Ollama)

For offline development and maximum privacy, you can run the models entirely on your local machine.

1. Ensure the Ollama daemon is running, then pull the necessary models:

ollama pull llama3    # Primary Model
ollama pull qwen2.5   # Fallback 1
ollama pull phi3      # Fallback 2

2. Launch the server with the local environment flag enabled:

USE_LOCAL_LLM=True uv run streamlit run app.py

---

☁️ Deploying to Streamlit Community Cloud

Deploying this application online for your club members is completely free and requires zero infrastructure maintenance.

1. Push your configured repository to GitHub.
2. Navigate to share.streamlit.io and click New app.
3. Select this repository and point the main file path to app.py.
4. Open the Advanced Settings -> Secrets menu and insert your Groq API Key:

GROQ_API_KEY = "gsk_your_api_key_here"

5. Click Deploy. The environment markers in pyproject.toml will automatically handle Linux compatibility constraints, yielding a fully functional public web app.

---

Built with ❤️ by Shreyas
Empowering the SRM Insider Community through AI