RagDocs

RagDocs is a Retrieval-Augmented Generation (RAG) platform for document-centric querying, semantic retrieval, and contextual AI inference. It provides a unified workspace for ingesting, indexing, and interacting with structured and unstructured datasets, including PDFs and CSV files.

The system combines vector search, embedding pipelines, and LLM-based reasoning to generate context-grounded responses with citation tracing.

Core Features

• Document-Centric RAG Pipeline
  Supports ingestion of PDF and CSV documents. Uploaded files are parsed, segmented into semantic chunks, embedded using transformer-based embedding models, and indexed in Qdrant for low-latency vector retrieval.

• Context-Grounded Response Generation
  Integrates OpenRouter and LangChain to generate responses constrained to retrieved document context. Outputs include source attribution for traceability and hallucination reduction.

• Semantic Retrieval Layer
  Uses dense vector similarity search through Qdrant to retrieve high-relevance context segments for downstream LLM inference.

• Corrective RAG Enhancements
  Improves retrieval quality using query rewriting, multi-query expansion, HyDE, frequency-based re-ranking, and LLM-based relevance judging.

• Session-Aware Conversational Interface
  Maintains conversational continuity while prioritizing document exploration and query resolution workflows.

• Cross-Platform Runtime
  Built with Expo and React Native, enabling deployment across Web, Android, and iOS from a shared codebase.

• Interactive Visualization Layer
  Includes graph-based knowledge visualization, animated UI transitions, and structured navigation components for document interaction workflows.

---

System Architecture

Frontend

Built using React Native and Expo with a modular component-driven architecture.

Stack

• Framework: Expo, Expo Router
• UI Layer: React Native
• Styling System: NativeWind
• Animation Runtime: React Native Reanimated
• Icon System: Lucide React Native
• Networking: Axios

Responsibilities

• File upload interface
• Conversational query interface
• Citation rendering
• Knowledge graph visualization
• Cross-platform state management and navigation

---

Backend

Node.js-based backend responsible for ingestion, embedding generation, retrieval orchestration, and inference routing.

Stack

• Runtime Framework: Express.js
• Vector Database: Qdrant
  (@qdrant/js-client-rest, @langchain/qdrant)
• LLM Gateway: OpenRouter
• Embedding Provider: OpenAI Embeddings
  (@langchain/openai)
• Document Parsing: pdf-parse, csv-parse
• Orchestration Layer: LangChain

Responsibilities

• Document parsing and preprocessing
• Text chunking and segmentation
• Embedding generation
• Vector indexing and retrieval
• Retrieval-augmented prompt construction
• Citation-aware response generation
• Query rewriting, multi-query expansion, and relevance judging for corrective RAG

---

Processing Pipeline

1. Document upload
2. Content extraction and parsing
3. Semantic chunk generation
4. Embedding vector creation
5. Qdrant vector indexing
6. Similarity-based retrieval
7. Context injection into LLM prompts
8. Citation-grounded response generation
9. Corrective RAG loop (rewrite, multi-query expansion, re-rank, and judge)

---

Installation

Prerequisites

• Node.js v18+
• A running Qdrant instance (local or cloud)
• OpenRouter API credentials

---

Backend Setup

Navigate to the backend directory:

cd Backend

Install dependencies:

npm install

Create a .env file inside Backend:

PORT=5000

QDRANT_URL=your_qdrant_cluster_url
QDRANT_API_KEY=your_qdrant_api_key

OPENROUTER_API_KEY=your_openrouter_api_key
MODEL=your_preferred_openrouter_model

Example model:

MODEL=openai/gpt-4o-mini

Start the development server:

npm run dev

Backend service will run at:

http://localhost:5000

---

Frontend Setup

Navigate to the frontend directory:

cd Frontend

Install dependencies:

npm install

Create a .env file inside Frontend:

EXPO_PUBLIC_API_URL=http://localhost:5000

Start the Expo development server:

npm start

Platform targets:

• Web: w
• iOS Simulator: i
• Android Emulator: a

---

Usage Workflow

1. Document Ingestion

Upload a PDF or CSV file through the frontend interface.

You can also index a public URL via the backend endpoint:

POST /api/upload/url
{
  "url": "https://example.com/your-article"
}

The backend performs:

• File parsing
• Semantic chunking
• Embedding generation
• Vector indexing into Qdrant

---

2. Retrieval-Augmented Querying

Submit natural language queries through the chat interface.

The system:

• Retrieves semantically relevant chunks
• Constructs a context-grounded prompt
• Generates responses using the configured LLM
• Returns citation-linked outputs derived from retrieved context

---

Supported File Types

• PDF
• CSV
• Web URLs (basic HTML text extraction)

---

Deployment Targets

• Web
• Android
• iOS