deploy: v2 production baked index (zero latency)

.dockerignore

@@ -0,0 +1,56 @@
+# Python
+__pycache__/
+*.pyc
+*.pyo
+*.pyd
+.Python
+
+# Virtual Environment
+venv/
+env/
+ENV/
+.venv
+
+# Environment Variables
+.env
+.env.local
+
+# Data directories
+data/raw/
+# data/index/ <-- We want to include this for the showcase
+data/db/
+data/processed/
+
+# IDE
+.vscode/
+.idea/
+*.swp
+*.swo
+*~
+
+# OS
+.DS_Store
+Thumbs.db
+
+# Git
+.git/
+.gitignore
+
+# Testing
+.pytest_cache/
+.coverage
+htmlcov/
+
+# Documentation (not needed in container)
+docs/
+*.md
+!README.md
+
+# Notebooks
+notebooks/
+*.ipynb
+
+# Build artifacts
+dist/
+build/
+*.egg-info/

.env.example

@@ -0,0 +1,6 @@
+OPENAI_API_KEY=sk-xxx
+VLLM_API_URL=http://localhost:8000/v1
+# VLLM_MODEL=mistralai/Mistral-7B-Instruct-v0.2
+LOG_LEVEL=INFO
+DATA_DIR=data
+INDEX_DIR=data/index

.github/workflows/deploy_to_hf.yml

@@ -0,0 +1,20 @@
+name: Sync to Hugging Face Hub
+
+on:
+  push:
+    branches: [main]
+
+jobs:
+  sync-to-hub:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v3
+        with:
+          fetch-depth: 0
+          lfs: true
+      - name: Push to hub
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN }}
+        run: |
+          git remote add hf https://yuvis:$HF_TOKEN@huggingface.co/spaces/yuvis/Enterprise-RAG-System
+          git push -f hf main

.github/workflows/docker-build.yml

@@ -0,0 +1,34 @@
+name: Build and Push Docker Image
+
+on:
+  push:
+    branches: [ "main", "master" ]
+  workflow_dispatch:
+
+jobs:
+  build-and-push:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Login to Docker Hub
+        uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
+
+      - name: Build and push
+        uses: docker/build-push-action@v5
+        with:
+          context: .
+          push: true
+          tags: ${{ secrets.DOCKERHUB_USERNAME }}/enterprise-rag:latest
+          cache-from: type=gha
+          cache-to: type=gha,mode=max

.gitignore

@@ -0,0 +1,23 @@
+# Python
+__pycache__/
+*.pyc
+
+# Virtual Env
+venv/
+env/
+
+# Environment Variables
+.env
+
+# System
+.DS_Store
+
+# Data (Generated/Downloaded)
+data/index/
+data/raw/
+# data/db/
+# data/processed/
+
+# IDE
+.vscode/
+.idea/

AWS_APP_RUNNER_SETUP.md

@@ -0,0 +1,48 @@
+# AWS App Runner Deployment Guide
+
+Follow these steps to deploy your Enterprise RAG System to AWS App Runner for a recruiter-ready showcase.
+
+## 1. Local Verification
+First, build and run your image locally to ensure the index is properly packaged:
+```bash
+docker build -t enterprise-rag .
+docker run -p 8501:8501 -e GROQ_API_KEY=your_key_here enterprise-rag
+```
+Visit `http://localhost:8501` to verify.
+
+## 2. Push to AWS ECR
+You need to push your image to the Amazon Elastic Container Registry.
+
+1. **Create Repository**:
+   ```bash
+   aws ecr create-repository --repository-name enterprise-rag --region your-region
+   ```
+2. **Login to ECR**:
+   ```bash
+   aws ecr get-login-password --region your-region | docker login --username AWS --password-stdin <your-account-id>.dkr.ecr.<your-region>.amazonaws.com
+   ```
+3. **Tag & Push**:
+   ```bash
+   docker tag enterprise-rag:latest <your-account-id>.dkr.ecr.<your-region>.amazonaws.com/enterprise-rag:latest
+   docker push <your-account-id>.dkr.ecr.<your-region>.amazonaws.com/enterprise-rag:latest
+   ```
+
+## 3. Create App Runner Service
+1. Go to **AWS Console** → **App Runner**.
+2. Click **Create service**.
+3. **Source**:
+   - Repository type: **Container registry**.
+   - Provider: **Amazon ECR**.
+   - Container image: Select your `enterprise-rag` image.
+   - Deployment settings: **Manual** (or Automatic if you want CI/CD).
+4. **Configuration**:
+   - Service name: `enterprise-rag-showcase`.
+   - Virtual CPU & Memory: **1 vCPU & 2 GB** (Minimum recommended).
+   - **Environment variables**:
+     - `GROQ_API_KEY`: Paste your key here.
+5. **Connectivity**:
+   - Port: **8501**.
+6. **Review & Create**.
+
+## 4. Final Result
+Once deployed, AWS will provide a public URL like `https://xxxxxx.us-east-1.awsapprunner.com`. This is the URL you can share with recruiters!

AWS_DEPLOYMENT.md

@@ -0,0 +1,152 @@
+# AWS Deployment Guide - Enterprise RAG System
+
+## Prerequisites
+- AWS Account
+- AWS CLI installed and configured
+- Docker installed locally
+
+## Deployment Options
+
+### Option 1: AWS EC2 (Recommended for Full Control)
+
+#### Step 1: Launch EC2 Instance
+1. Go to AWS Console → EC2
+2. Click **"Launch Instance"**
+3. Choose:
+   - **AMI**: Ubuntu 22.04 LTS
+   - **Instance Type**: t3.medium (4GB RAM minimum)
+   - **Storage**: 20GB
+4. Configure Security Group:
+   - Allow SSH (port 22) from your IP
+   - Allow HTTP (port 8501) from anywhere
+   - Allow HTTP (port 8000) from anywhere
+
+#### Step 2: Connect to Instance
+```bash
+ssh -i your-key.pem ubuntu@your-ec2-public-ip
+```
+
+#### Step 3: Install Docker
+```bash
+# Update system
+sudo apt update && sudo apt upgrade -y
+
+# Install Docker
+curl -fsSL https://get.docker.com -o get-docker.sh
+sudo sh get-docker.sh
+sudo usermod -aG docker ubuntu
+
+# Install Docker Compose
+sudo apt install docker-compose-plugin
+```
+
+#### Step 4: Clone Repository
+```bash
+git clone https://github.com/YuvrajSinghBhadoria2/Enterprise-RAG-System.git
+cd Enterprise-RAG-System
+```
+
+#### Step 5: Configure Environment
+```bash
+# Create .env file
+cat > .env << EOF
+GROQ_API_KEY=your_groq_api_key_here
+EOF
+```
+
+#### Step 6: Build and Run
+```bash
+# Using Docker Compose
+docker compose -f docker/docker-compose.yml up -d --build
+
+# Generate data (one-time)
+docker compose -f docker/docker-compose.yml exec api python3 tools/generate-dataset.py
+docker compose -f docker/docker-compose.yml exec api python3 src/ingestion/ingest.py
+```
+
+#### Step 7: Access Application
+- **UI**: `http://your-ec2-public-ip:8501`
+- **API**: `http://your-ec2-public-ip:8000/docs`
+
+---
+
+### Option 2: AWS ECS (Fargate) - Serverless
+
+#### Step 1: Push Docker Image to ECR
+```bash
+# Create ECR repository
+aws ecr create-repository --repository-name enterprise-rag
+
+# Login to ECR
+aws ecr get-login-password --region us-east-1 | docker login --username AWS --password-stdin YOUR_ACCOUNT_ID.dkr.ecr.us-east-1.amazonaws.com
+
+# Build and push
+docker build -f docker/Dockerfile.api -t enterprise-rag .
+docker tag enterprise-rag:latest YOUR_ACCOUNT_ID.dkr.ecr.us-east-1.amazonaws.com/enterprise-rag:latest
+docker push YOUR_ACCOUNT_ID.dkr.ecr.us-east-1.amazonaws.com/enterprise-rag:latest
+```
+
+#### Step 2: Create ECS Task Definition
+1. Go to ECS Console
+2. Create new Task Definition (Fargate)
+3. Add container:
+   - Image: Your ECR image URI
+   - Memory: 4GB
+   - Port: 8501
+4. Add environment variable: `GROQ_API_KEY`
+
+#### Step 3: Create ECS Service
+1. Create ECS Cluster
+2. Create Service from Task Definition
+3. Configure Load Balancer (optional)
+
+---
+
+### Option 3: AWS Lightsail (Simplest)
+
+#### Step 1: Create Lightsail Instance
+1. Go to Lightsail Console
+2. Create Instance:
+   - Platform: Linux/Unix
+   - Blueprint: Ubuntu 22.04
+   - Plan: $10/month (2GB RAM)
+
+#### Step 2: Deploy
+Same as EC2 steps 2-7 above
+
+---
+
+## Cost Estimates
+
+| Service | Cost/Month | Best For |
+|---------|-----------|----------|
+| EC2 t3.medium | ~$30 | Full control, testing |
+| ECS Fargate | ~$40 | Production, auto-scaling |
+| Lightsail | $10-20 | Simple deployment |
+
+## Recommended: EC2 t3.medium
+
+For your use case, I recommend **EC2 t3.medium** because:
+- ✅ Full control
+- ✅ Easy to manage
+- ✅ Cost-effective
+- ✅ Can run Docker Compose easily
+
+## Maintenance
+
+**Update code:**
+```bash
+cd Enterprise-RAG-System
+git pull
+docker compose -f docker/docker-compose.yml up -d --build
+```
+
+**View logs:**
+```bash
+docker compose -f docker/docker-compose.yml logs -f
+```
+
+**Restart services:**
+```bash
+docker compose -f docker/docker-compose.yml restart
+```

DEPLOYMENT.md

@@ -0,0 +1,178 @@
+# Deployment Guide
+
+ This guide ensures a smooth deployment of the Enterprise RAG system to any cloud VPS (Virtual Private Server) such as AWS EC2, DigitalOcean Droplet, Google Compute Engine, or Azure VM.
+
+## 🚀 Prerequisites
+
+*   **Server**: A Linux server (Ubuntu 22.04 LTS recommended).
+*   **Specs**: Minimum 4GB RAM (8GB recommended for embeddings/FAISS), 2 vCPUs.
+*   **Software**: Docker and Docker Compose installed.
+
+## � Recommended Providers
+
+ **Railway.app** (Easiest PaaS):
+    *   Perfect for quick demos.
+    *   Supports our `Dockerfile` setup out of the box.
+
+## 🚂 Railway.app Deployment (Complete Setup)
+
+Railway requires deploying **two separate services** from the same repository.
+
+### Prerequisites
+*   GitHub repository: `https://github.com/YuvrajSinghBhadoria2/Enterprise-RAG-System.git`
+*   Railway account: [railway.app](https://railway.app)
+*   Groq API Key
+
+### Step 1: Deploy API Service
+
+1. **Create New Project** in Railway
+2. **Deploy from GitHub** → Select your repository
+3. Railway will auto-detect the `railway.toml` and use the Dockerfile
+4. **Add Environment Variables**:
+   ```
+   GROQ_API_KEY=gsk_your_key_here
+   PORT=8000
+   ```
+5. **Deploy** - Railway will build using `docker/Dockerfile.api`
+6. **Get API URL** - Copy the public URL (e.g., `https://enterprise-rag-production.up.railway.app`)
+
+### Step 2: Deploy UI Service
+
+1. In the **same Railway project**, click **+ New Service**
+2. **Deploy from GitHub** → Select the **same repository**
+3. **Configure Build**:
+   - Go to **Settings** → **Build**
+   - Set **Dockerfile Path**: `docker/Dockerfile.streamlit`
+4. **Add Environment Variable**:
+   ```
+   API_URL=https://your-api-url-from-step1.up.railway.app/api/v1/chat
+   ```
+   *(Replace with your actual API URL from Step 1)*
+5. **Deploy**
+
+### Step 3: Generate Data (Critical!)
+
+Railway containers are ephemeral, so you need to generate data on startup:
+
+**Option A: One-time manual generation** (for testing):
+```bash
+# In Railway API service shell
+python3 tools/generate-dataset.py
+python3 src/ingestion/ingest.py
+```
+
+**Option B: Auto-generate on startup** (recommended):
+Update the `railway.toml` start command to include data generation.
+
+### Step 4: Access Your Application
+
+- **UI**: `https://your-ui-service.up.railway.app`
+- **API Docs**: `https://your-api-service.up.railway.app/docs`
+
+### Troubleshooting
+
+**Build Timeout?**
+- Ensure `.dockerignore` excludes `venv/` and `data/`
+- Check that `railway.toml` points to the correct Dockerfile
+
+**UI Can't Connect to API?**
+- Verify `API_URL` environment variable in UI service
+- Ensure API service is deployed and running
+- Check API URL includes `/api/v1/chat` endpoint
+
+---
+
+## 📦 Step-by-Step Deployment (VPS)
+
+### 1. Provision & Access Server
+SSH into your server:
+```bash
+ssh user@your-server-ip
+```
+
+### 2. Install Docker (If not installed)
+```bash
+# Update packages
+sudo apt update && sudo apt upgrade -y
+
+# Install Docker
+curl -fsSL https://get.docker.com -o get-docker.sh
+sudo sh get-docker.sh
+
+# Install Docker Compose Plugin
+sudo apt install docker-compose-plugin
+```
+
+### 3. Clone the Repository
+```bash
+git clone https://github.com/your-repo/enterprise-rag.git
+cd enterprise-rag
+```
+
+### 4. Configure Environment
+Create the production `.env` file:
+```bash
+cp .env.example .env
+nano .env
+```
+*Paste your `GROQ_API_KEY` or `OPENAI_API_KEY` into the file.*
+
+### 5. Build and Start Services
+This command will build the images and start the API and UI in the background with auto-restart enabled.
+```bash
+# Using the Makefile shortcut
+make up
+
+# OR manually using docker compose
+docker compose -f docker/docker-compose.yml up -d --build
+```
+
+### 6. Generate Data (Critical Step)
+Fresh deployments start empty. You must ingest the datasets to make the search work.
+```bash
+# Run ingestion inside the running API container
+docker compose -f docker/docker-compose.yml exec api python3 tools/generate-dataset.py
+docker compose -f docker/docker-compose.yml exec api python3 src/ingestion/ingest.py
+```
+
+### 7. Access the Application
+*   **UI**: `http://your-server-ip:8501`
+*   **API**: `http://your-server-ip:8000/docs`
+
+---
+
+## 🔒 Production Hardening
+
+### 1. Firewall (UFW)
+Only open necessary ports.
+```bash
+sudo ufw allow 22/tcp
+sudo ufw allow 8501/tcp  # Streamlit
+sudo ufw allow 8000/tcp  # API
+sudo ufw enable
+```
+
+### 2. Reverse Proxy (Nginx + SSL)
+For HTTPS, use Nginx as a reverse proxy.
+```nginx
+server {
+    listen 80;
+    server_name rag.yourdomain.com;
+
+    location / {
+        proxy_pass http://localhost:8501;
+        proxy_set_header Host $host;
+        proxy_set_header X-Real-IP $remote_addr;
+    }
+}
+```
+
+## 🛠️ Maintenance
+
+*   **View Logs**: `make logs`
+*   **Restart Services**: `make down && make up`
+*   **Update Code**:
+    ```bash
+    git pull
+    make up
+    ```

Dockerfile

@@ -0,0 +1,27 @@
+FROM python:3.11-slim
+
+WORKDIR /app
+
+# Install system dependencies
+RUN apt-get update && apt-get install -y \
+    build-essential \
+    && rm -rf /var/lib/apt/lists/*
+
+# Copy requirements and install Python dependencies
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Copy application code
+COPY . .
+
+# Download index during build (Bake into image)
+RUN python tools/download_index.py
+
+# Set Python path
+ENV PYTHONPATH=/app
+
+# Expose Streamlit port
+EXPOSE 8501
+
+# Run Streamlit
+CMD ["streamlit", "run", "app.py", "--server.port=8501", "--server.address=0.0.0.0"]

Makefile

@@ -0,0 +1,40 @@
+.PHONY: build up down logs ingest eval run-local
+
+# Docker commands
+build:
+	docker-compose -f docker/docker-compose.yml --env-file .env build
+
+up:
+	docker-compose -f docker/docker-compose.yml --env-file .env up -d --build
+
+down:
+	docker-compose -f docker/docker-compose.yml --env-file .env down
+
+logs:
+	docker-compose -f docker/docker-compose.yml --env-file .env logs -f
+
+api-shell:
+	docker-compose -f docker/docker-compose.yml --env-file .env exec api /bin/bash
+
+# Run evaluation inside Docker
+eval:
+	docker-compose -f docker/docker-compose.yml --env-file .env exec api python3 tools/run_eval.py
+
+# Run evaluation locally (Mac fallback)
+eval-local:
+	export DISABLE_FAISS=1 && export KMP_DUPLICATE_LIB_OK=TRUE && export GROQ_API_KEY=${GROQ_API_KEY} && python3 tools/run_eval.py
+
+# Ingestion (runs locally if venv active, or use via docker exec)
+ingest:
+	export PYTHONPATH=$$PYTHONPATH:. && python3 src/ingestion/ingest.py
+
+# Data generation
+generate-data:
+	python3 tools/generate-dataset.py
+
+# Run API and UI locally (Mac fallback)
+run-local:
+	@echo "Starting Enterprise RAG Locally (Safe Mode)..."
+	@export DISABLE_FAISS=1 && export KMP_DUPLICATE_LIB_OK=TRUE && export GROQ_API_KEY=${GROQ_API_KEY} && \
+	(uvicorn src.app.main:app --host 0.0.0.0 --port 8000 &) && \
+	(sleep 5 && streamlit run src/ui/app.py --server.port 8501)

Procfile

@@ -0,0 +1 @@
+web: uvicorn src.app.main:app --host 0.0.0.0 --port $PORT

README.md

@@ -0,0 +1,170 @@
+---
+title: Enterprise RAG System
+emoji: 🚀
+colorFrom: blue
+colorTo: indigo
+sdk: docker
+app_port: 8501
+pinned: false
+---
+
+# Enterprise RAG System
+
+An enterprise-grade Retrieval-Augmented Generation (RAG) system designed for high accuracy, safety, and scalability. This project demonstrates a production-ready pipeline with hybrid search, reranking, and strict guardrails against hallucinations.
+
+## 🚀 Key Features
+
+*   **Hybrid Retrieval**: Combines **BM25** (Keyword) and **FAISS** (Dense Vector) search for optimal recall.
+*   **Context Reranking**: Utilizes `cross-encoder/ms-marco-MiniLM-L-6-v2` to precision-rank documents before generation.
+*   **Enterprise Guardrails**:
+    *   **Refusal Logic**: Strictly refuses to answer if context is insufficient.
+    *   **Halucination Detection**: Automated grading of Answer Relevancy and Groundedness.
+    *   **Confidence Gating**: Blocks generation if retrieval scores are below a safety threshold.
+*   **Multi-Provider LLM**: Supports **Groq** (Llama-3), **vLLM**, and **OpenAI**.
+*   **Modern Stack**: Built with **FastAPI**, **Streamlit**, and **Docker**.
+
+---
+
+## 🛠️ Quick Start
+
+### Prerequisites
+*   Docker Desktop (Recommended)
+*   Python 3.10+ (For local run)
+*   Groq API Key (or OpenAI/vLLM)
+
+### 1. Configuration
+Create a `.env` file in the root directory:
+```bash
+cp .env.example .env
+```
+Edit `.env` and add your API key:
+```ini
+GROQ_API_KEY=gsk_...
+# Cloud Vector DB (Optional - Recommended for Deployment)
+VECTOR_DB_TYPE=pinecone
+PINECONE_API_KEY=pcsk_...
+
+# Optional:
+OPENAI_API_KEY=sk-...
+```
+
+### 2. Generate Data
+The system needs data to function. Run the ingestion script to download real datasets (WikiQA, Multi-News, GovReport) and build the index:
+```bash
+make generate-data
+make ingest
+```
+
+### 3. Run the Application
+
+#### Option A: Docker (Recommended for Linux/Windows)
+The most stable environment.
+```bash
+make up
+```
+*   **UI**: [http://localhost:8501](http://localhost:8501)
+*   **API**: [http://localhost:8000/docs](http://localhost:8000/docs)
+
+#### Option B: Local Safe Mode (Apple Silicon / Mac)
+Use this if you encounter Docker connection issues or `Segmentation Fault` (FAISS/OpenMP conflicts). This mode disables FAISS and runs on **BM25 (Keyword Search) only**, ensuring stability.
+```bash
+make run-local
+```
+*(This starts both the FastAPI backend and Streamlit UI)*
+
+---
+
+## 📊 Evaluation
+
+Verify the accuracy and safety of the system using the built-in evaluation suite.
+
+### Run Evaluation (Local Safe Mode)
+```bash
+make eval-local
+```
+This script will:
+1.  Load the **WikiQA** test set.
+2.  Run the full RAG pipeline for each question.
+3.  Report:
+    *   **Recall@10**: Retrieval effectiveness.
+    *   **MRR**: Mean Reciprocal Rank.
+    *   **Groundedness**: Frequency of hallucination checks passing.
+    *   **Refusal Rate**: How often the system correctly refuses unknown questions.
+
+## 🏆 Performance & Results
+
+Tested on **WikiQA**, **Multi-News**, and **GovReport** datasets.
+
+| Metric | Score | Description |
+| :--- | :--- | :--- |
+| **Recall@10** | **1.0000** | Perfect retrieval of relevant documents. |
+| **MRR** | **1.0000** | Relevant document consistently ranked #1. |
+| **Factuality** | **1.0000** | 100% of answers grounded in context. |
+| **Safety** | **100%** | Successfully refuses to answer out-of-context queries. |
+
+### 🛡️ Guardrails in Action
+
+**1. Hallucination Prevention (Safety Layer)**
+![Hallucination Blocked](docs/assets/refusal_screenshot.png)
+*   **Scenario Refusal**: The system correctly identified that the retrieved context (about Howard Stern) was irrelevant to the "Airline Strike" query.
+*   **Low Confidence**: The retrieval score of `-9.73` triggered the safety gate (Threshold: `-4.0`), automatically blocking the generation.
+*   **Result**: Zero hallucination. The user receives a safe, honest refusal instead of a made-up answer.
+
+**2. Enterprise Accuracy**
+![Detailed Answer](docs/assets/success_screenshot.png)
+*   **High Precision**: The query for "Emerging Contaminants" retrieved exact matches from the *GovReport* dataset.
+*   **Grounded Generation**: The answer is derived *strictly* from the text, listing specific chemicals (Perchlorate, TCE, etc.) mentioned in the document.
+*   **Verified**: Retrieval score of `4.82` shows high confidence, allowing the answer to pass.
+
+---
+
+## 🏗️ Architecture
+
+```mermaid
+flowchart TD
+    User([User Query]) --> Hybrid{Hybrid Retrieval}
+    
+    subgraph Retrieval Layer
+        Hybrid -->|Lexical| BM25[BM25 Index]
+        Hybrid -->|Semantic| FAISS[FAISS Vector DB]
+        BM25 --> Candidates[Candidate Pool]
+        FAISS --> Candidates
+    end
+    
+    Candidates --> Rerank[Cross-Encoder Reranker]
+    
+    subgraph Safety Layer
+        Rerank --> Gate{Confidence Gate}
+        Gate -- Score < -4.0 --> Refusal([Refusal Response])
+        Gate -- Score >= -4.0 --> Context[Context Optimization]
+    end
+    
+    subgraph Generation Layer
+        Context --> Prompt[Strict System Prompt]
+        Prompt --> LLM[LLM Inference]
+        LLM --> Guard[Hallucination Check]
+        Guard --> Final([Final Answer])
+    end
+
+    style Refusal fill:#ffcccc,stroke:#333
+    style Final fill:#ccffcc,stroke:#333
+    style Gate fill:#fff4e6,stroke:#ff9900
+```
+
+1.  **Ingestion**: Documents are cleaned, chunked (Sliding Window), and indexed into **Faiss** (Vectors) and **BM25** (Keywords).
+2.  **Retrieval**: Queries retrieve candidates from both indices (`HybridRetriever`).
+3.  **Reranking**: A Cross-Encoder scores the relevance of each candidate pair (`Query, Doc`).
+4.  **Guardrails**:
+    *   If `Max(Rerank Score) < -4.0`: **Refuse** immediately.
+5.  **Generation**: Top documents are passed to the LLM with a strict "Context-Only" system prompt.
+6.  **Validation**: Output is graded for Groundedness (Token Overlap) before being returned (in Evaluation mode).
+
+## 📁 Project Structure
+
+*   `src/app`: FastAPI Backend
+*   `src/ui`: Streamlit Frontend
+*   `src/pipeline`: Core RAG Logic (`QueryPipeline.py`)
+*   `src/retriever`: Search Algorithms (`HybridRetriever.py`)
+*   `src/eval`: Scoring Metrics (`Hallucination`, `Relevancy`)
+*   `tools`: Scripts for dataset generation and evaluation
+*   `data`: Raw and Indexed data storage

app.py

@@ -0,0 +1,173 @@
+"""
+Hugging Face Spaces - Enterprise RAG System
+Standalone Streamlit application
+"""
+
+import streamlit as st
+import os
+import sys
+
+
+
+# Add src to path
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), 'src'))
+
+from dotenv import load_dotenv
+load_dotenv()
+
+from src.pipeline.query_pipeline import QueryPipeline
+from src.ingestion.ingest import IngestionPipeline
+import subprocess
+
+def prepare_data():
+    """Ensure data is generated and indexed if missing"""
+    INDEX_DIR = "data/index"
+    RAW_DIR = "data/raw"
+    
+    # 1. Create directories
+    os.makedirs(INDEX_DIR, exist_ok=True)
+    os.makedirs(RAW_DIR, exist_ok=True)
+    
+    # 2. Check if raw data exists (Only needed if NOT using Pinecone)
+    if os.getenv("VECTOR_DB_TYPE", "").lower() != "pinecone":
+        if not os.listdir(RAW_DIR):
+            st.error("❌ Data folder empty! Please commit your 'data/raw' folder to Git and redeploy.")
+            st.stop()
+
+    # 3. Check if indices exist, if not run ingestion (Skip for Pinecone)
+    # 3. Check if indices exist (Files should be baked in)
+    bm25_path = os.path.join(INDEX_DIR, "bm25.pkl")
+    
+    # Only download if absolutely missing (Fallback for dev env)
+    if not os.path.exists(bm25_path):
+        with st.spinner("Downloading Knowledge Base (Dev Mode)..."):
+             try:
+                from huggingface_hub import hf_hub_download
+                os.makedirs(INDEX_DIR, exist_ok=True)
+                hf_hub_download(repo_id="yuvis/enterprise-rag-index", filename="index/bm25.pkl", repo_type="dataset", local_dir="data")
+                hf_hub_download(repo_id="yuvis/enterprise-rag-index", filename="index/doc_map.pkl", repo_type="dataset", local_dir="data")
+             except Exception:
+                 pass
+
+st.set_page_config(
+    page_title="Enterprise RAG Search",
+    page_icon="🔍",
+    layout="wide"
+)
+
+# Initialize pipeline
+@st.cache_resource
+def load_pipeline():
+    """Load the RAG pipeline (cached for performance)"""
+    try:
+        # Ensure data is ready before initializing pipeline
+        prepare_data()
+        return QueryPipeline()
+    except Exception as e:
+        st.error(f"Error loading pipeline: {e}")
+        st.exception(e)
+        return None
+
+# Main UI
+st.title("🔍 Enterprise RAG Search")
+st.markdown("*Production-grade Retrieval-Augmented Generation with Hallucination Prevention*")
+
+# Sidebar configuration
+with st.sidebar:
+    st.header("⚙️ Configuration")
+    st.caption("🚀 Version: Pinecone V2")
+    
+    # Check for API key
+    groq_key = os.getenv("GROQ_API_KEY")
+    if not groq_key:
+        st.warning("⚠️ GROQ_API_KEY not set. Please configure in Space settings.")
+    else:
+        st.success("✅ API Key configured")
+    
+    st.divider()
+    
+    top_k_retrieval = st.slider("Retrieval Top-K", 5, 50, 20)
+    top_k_rerank = st.slider("Rerank Top-K", 1, 10, 5)
+    
+    st.divider()
+    st.markdown("### 📊 System Info")
+    st.info("""
+    - **Hybrid Search**: BM25 + FAISS
+    - **Reranking**: Cross-Encoder
+    - **Safety**: Confidence Gating
+    """)
+
+# Initialize session state
+if "messages" not in st.session_state:
+    st.session_state.messages = []
+
+# Display chat history
+for message in st.session_state.messages:
+    with st.chat_message(message["role"]):
+        st.markdown(message["content"])
+
+# Chat input
+if prompt := st.chat_input("Ask a question about your documents..."):
+    # Add user message
+    st.session_state.messages.append({"role": "user", "content": prompt})
+    with st.chat_message("user"):
+        st.markdown(prompt)
+    
+    # Generate response
+    with st.chat_message("assistant"):
+        with st.spinner("Searching and generating answer..."):
+            pipeline = load_pipeline()
+            
+            if pipeline is None:
+                st.error("Pipeline not loaded. Please check configuration.")
+            else:
+                try:
+                    result = pipeline.run(
+                        query=prompt,
+                        top_k_retrieval=top_k_retrieval,
+                        top_k_rerank=top_k_rerank
+                    )
+                    
+                    # Display answer
+                    st.markdown(result["answer"])
+                    
+                    # Display metadata in expander
+                    with st.expander("📋 View Details"):
+                        col1, col2, col3 = st.columns(3)
+                        
+                        with col1:
+                            st.metric("Retrieval Score", f"{result.get('retrieval_score', 'N/A'):.2f}")
+                        
+                        with col2:
+                            hallucination = result.get('hallucination_score', 'N/A')
+                            if hallucination != 'N/A':
+                                st.metric("Hallucination Score", f"{hallucination:.2f}")
+                        
+                        with col3:
+                            groundedness = result.get('groundedness', 'N/A')
+                            if groundedness != 'N/A':
+                                st.metric("Groundedness", f"{groundedness:.2f}")
+                        
+                        # Show retrieved context
+                        if result.get("context"):
+                            st.markdown("**Retrieved Context:**")
+                            for i, (doc, score) in enumerate(result["context"][:3], 1):
+                                st.markdown(f"{i}. [Score: {score:.2f}] {doc[:200]}...")
+                    
+                    # Add to chat history
+                    st.session_state.messages.append({
+                        "role": "assistant",
+                        "content": result["answer"]
+                    })
+                    
+                except Exception as e:
+                    st.error(f"Error generating response: {e}")
+                    st.exception(e)
+
+# Footer
+st.divider()
+st.markdown("""
+<div style='text-align: center; color: gray; font-size: 0.8em;'>
+    Enterprise RAG System | <a href='https://github.com/YuvrajSinghBhadoria2/Enterprise-RAG-System'>GitHub</a>
+</div>
+""", unsafe_allow_html=True)

configs/default.yaml

@@ -0,0 +1,22 @@
+app:
+  title: "Enterprise RAG Search"
+  host: "0.0.0.0"
+  port: 8080
+
+retrieval:
+  top_k_retrieval: 20
+  top_k_rerank: 5
+  weights:
+    bm25: 0.3
+    dense: 0.7
+
+embeddings:
+  model_name: "BAAI/bge-m3"
+  device: "cpu" # or cuda
+
+reranker:
+  model_name: "cross-encoder/ms-marco-MiniLM-L-6-v2"
+
+ingestion:
+  chunk_size: 512
+  chunk_overlap: 50

data/raw/finphrase_000.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+According to Gran , the company has no plans to move all production to Russia , although that is where the company is growing .

data/raw/finphrase_001.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+For the last quarter of 2010 , Componenta 's net sales doubled to EUR131m from EUR76m for the same period a year earlier , while it moved to a zero pre-tax profit from a pre-tax loss of EUR7m .

data/raw/finphrase_002.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+In the third quarter of 2010 , net sales increased by 5.2 % to EUR 205.5 mn , and operating profit by 34.9 % to EUR 23.5 mn .

data/raw/finphrase_003.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+Operating profit rose to EUR 13.1 mn from EUR 8.7 mn in the corresponding period in 2007 representing 7.7 % of net sales .

data/raw/finphrase_004.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+Operating profit totalled EUR 21.1 mn , up from EUR 18.6 mn in 2007 , representing 9.7 % of net sales .

data/raw/finphrase_005.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+Finnish Talentum reports its operating profit increased to EUR 20.5 mn in 2005 from EUR 9.3 mn in 2004 , and net sales totaled EUR 103.3 mn , up from EUR 96.4 mn .

data/raw/finphrase_006.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+Clothing retail chain Sepp+ñl+ñ 's sales increased by 8 % to EUR 155.2 mn , and operating profit rose to EUR 31.1 mn from EUR 17.1 mn in 2004 .

data/raw/finphrase_007.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+Consolidated net sales increased 16 % to reach EUR74 .8 m , while operating profit amounted to EUR0 .9 m compared to a loss of EUR0 .7 m in the prior year period .

data/raw/finphrase_008.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+Foundries division reports its sales increased by 9.7 % to EUR 63.1 mn from EUR 57.5 mn in the corresponding period in 2006 , and sales of the Machine Shop division increased by 16.4 % to EUR 41.2 mn from EUR 35.4 mn in the corresponding period in 2006 .

data/raw/finphrase_009.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+HELSINKI ( AFX ) - Shares closed higher , led by Nokia after it announced plans to team up with Sanyo to manufacture 3G handsets , and by Nokian Tyres after its fourth-quarter earnings report beat analysts ' expectations , dealers said .

data/raw/finphrase_010.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+Its board of directors will propose a dividend of EUR0 .12 per share for 2010 , up from the EUR0 .08 per share paid in 2009 .

data/raw/finphrase_011.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+MegaFon 's subscriber base increased 16.1 % in 2009 to 50.5 million users as of December 31 , while its market share by the number of customers amounted to 24 % as of late 2009 , up from 23 % as of late 2008 , according to TeliaSonera estimates .

data/raw/finphrase_012.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+Net income from life insurance doubled to EUR 6.8 mn from EUR 3.2 mn , and net income from non-life insurance rose to EUR 5.2 mn from EUR 1.5 mn in the corresponding period in 2009 .

data/raw/finphrase_013.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+Net sales increased to EUR193 .3 m from EUR179 .9 m and pretax profit rose by 34.2 % to EUR43 .1 m. ( EUR1 = USD1 .4 )

data/raw/finphrase_014.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+Net sales surged by 18.5 % to EUR167 .8 m. Teleste said that EUR20 .4 m , or 12.2 % , of the sales came from the acquisitions made in 2009 .

data/raw/finphrase_015.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+Nordea Group 's operating profit increased in 2010 by 18 percent year-on-year to 3.64 billion euros and total revenue by 3 percent to 9.33 billion euros .

data/raw/finphrase_016.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+Operating profit for the nine-month period increased from EUR13 .6 m , while net sales increased from EUR394 .7 m , as compared to the corresponding period in 2005 .

data/raw/finphrase_017.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+Operating profit for the three-month period increased from EUR1 .2 m , while revenue increased from EUR20 .2 m , as compared to the corresponding period in 2005 .

data/raw/finphrase_018.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+The company 's net profit rose 11.4 % on the year to 82.2 million euros in 2005 on sales of 686.5 million euros , 13.8 % up on the year , the company said earlier .

data/raw/finphrase_019.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+The Lithuanian beer market made up 14.41 million liters in January , a rise of 0.8 percent from the year-earlier figure , the Lithuanian Brewers ' Association reporting citing the results from its members .

data/raw/finphrase_020.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+Viking Line 's cargo revenue increased by 5.4 % to EUR 21.46 mn , and cargo volume increased by 2.4 % to 70,116 cargo units .

data/raw/finphrase_021.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+The fair value of the property portfolio doubled as a result of the Kapiteeli acquisition and totalled EUR 2,686.2 1,259.7 million .

data/raw/finphrase_022.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+10 February 2011 - Finnish media company Sanoma Oyj HEL : SAA1V said yesterday its 2010 net profit almost tripled to EUR297 .3 m from EUR107 .1 m for 2009 and announced a proposal for a raised payout .

data/raw/finphrase_023.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+A Helsinki : ELIiV today reported EPS of EUR1 .13 for 2009 , an increase over EPS of EUR1 .12 in 2008 .

data/raw/finphrase_024.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+Commission income increased by 22 % to EUR 4.4 mn , and lending volume rose by 13.5 % .

data/raw/finphrase_025.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+In January , traffic , measured in revenue passenger kilometres RPK , went up by 3.2 % and capacity , measured in available seat kilometres ASK , rose by 12.2 % .

data/raw/finphrase_026.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+In January-September 2010 , Fiskars ' net profit went up by 14 % year-on-year to EUR 65.4 million and net sales to EUR 525.3 million from EUR 487.7 million .

data/raw/finphrase_027.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+Net income from life insurance rose to EUR 16.5 mn from EUR 14.0 mn , and net income from non-life insurance to EUR 22.6 mn from EUR 15.2 mn in 2009 .

data/raw/finphrase_028.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+Sales have risen in other export markets .

data/raw/finphrase_029.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+Sales increased due to growing market rates and increased operations .

data/raw/finphrase_030.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+The agreement strengthens our long-term partnership with Nokia Siemens Networks .

data/raw/finphrase_031.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+The company 's order book stood at 1.5 bln euro $ 2.2 bln on September 30 , 2007 , up by 24.2 pct on the year , with international orders amounting to 365 mln euro $ 534.3 mln .

data/raw/finphrase_032.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+The company said that paper demand increased in all of its main markets , including of publication papers , and that it increased average paper prices by 4 percent compared with last year .

data/raw/finphrase_033.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+The world 's second largest stainless steel maker said net profit in the three-month period until Dec. 31 surged to euro603 million US$ 781 million , or euro3 .33 US$ 4.31 per share , from euro172 million , or euro0 .94 per share , the previous year .

data/raw/finphrase_034.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+Shares of Standard Chartered ( STAN ) rose 1.2 % in the FTSE 100 , while Royal Bank of Scotland ( RBS ) shares rose 2 % and Barclays shares ( BARC ) ( BCS ) were up 1.7 % .

data/raw/finphrase_035.txt

@@ -0,0 +1,3 @@
+Source: FinancialPhrasebank
+
+Shares of Nokia Corp. rose Thursday after the cell phone maker said its third-quarter earnings almost doubled and its share of the global handset market increased .