DocLens AI

A Product-Focused Research Intelligence Platform

DocLens AI is a full-stack platform designed to transform static research papers into an interactive intelligence system. It provides researchers with powerful semantic search, high-accuracy Retrieval-Augmented Generation (RAG), and a direct line of sight from AI-generated answers back to the original source text.

DocLens focuses on delivering robust, grounded AI features without the hallucination risks or brittleness of complex, speculative pipelines.

---

🎯 Core Capabilities (What DocLens Actually Does)

1. Research Collections & Library: Organize your PDFs into targeted domains.
2. Fast & Reliable Ingestion: Upload PDFs and watch them get automatically extracted, chunked, and embedded into a high-speed FAISS vector index.
3. Citation-Backed Research Chat: Ask questions across your collections. DocLens uses an advanced RAG pipeline to retrieve relevant chunks and generate answers. Every claim is backed by a clickable citation that reveals the exact source paragraph.
4. Literature Reviews: Automatically synthesize multi-paper literature reviews based only on retrieved text. Export them to Markdown or PDF.
5. Insights & Analytics: Track your corpus size, processing health, and discovery velocity.
6. Workspace Notes: Keep track of manual synthesis and thoughts alongside your automated chats.

Note: DocLens strictly relies on retrieved paper content and does not invent information. We do not support "Knowledge Graphs", "Neo4j integrations", or "Auto-generated concept pipelines", as we prioritize grounded, verifiable answers over speculative extraction.

---

🏗️ System Architecture

DocLens operates as a monorepo containing a React frontend, a NestJS core API, and a Python-based FastAPI AI service.

Technology Stack

Frontend Layer

• Framework: React 18 powered by Vite
• Styling: Custom Vanilla CSS (Dark-theme, glassmorphism aesthetics)
• State Management: React Context API
• Icons: Lucide React

Core Backend Layer

• Framework: NestJS (TypeScript)
• Database: PostgreSQL
• ORM: Prisma
• Authentication: JWT (JSON Web Tokens)

AI & Semantic Service Layer

• Framework: FastAPI (Python)
• Vector Database: FAISS (Facebook AI Similarity Search)
• Embeddings: HuggingFace sentence-transformers/all-MiniLM-L6-v2
• LLM Provider: OpenRouter (DeepSeek V3 default, Gemini 2.5 Flash fallback)
• Document Parsing: PyMuPDF (fitz)

---

📁 Project Structure

doclens-fullstack/
├── frontend/                   # React SPA
│   ├── public/                 # Static assets
│   ├── src/
│   │   ├── components/         # Reusable UI components
│   │   ├── pages/              # Route-level components
│   │   ├── contexts/           # Application state providers
│   │   ├── styles/             # Application design system
│   │   └── services/           # API integration layer
│
├── backend/                    # NestJS Core API
│   ├── src/
│   │   ├── auth/               # Authentication & Authorization
│   │   ├── users/              # User management
│   │   ├── collections/        # Research collections
│   │   ├── documents/          # Document state machine
│   │   └── query/              # Proxy to AI Service
│   ├── prisma/                 # Database schema and migrations
│
├── ai-service/                 # FastAPI Python Semantic Engine
│   ├── main.py                 # REST Endpoints
│   ├── ingest.py               # PDF Parsing, Chunking & Embedding
│   ├── query.py                # FAISS Retrieval & LLM Generation
│   └── vector_store/           # Local FAISS index & metadata DB

---

🚀 Setup & Installation

Prerequisites

• Node.js (v18 or higher)
• Python (3.9 or higher)
• PostgreSQL (v14 or higher)

1. Database Configuration

Ensure PostgreSQL is running, then create the database:

createdb doclens

Configure the backend environment:

cd backend
cp .env.example .env

Update backend/.env with your credentials:

DATABASE_URL="postgresql://username:password@localhost:5432/doclens"
JWT_SECRET="your-secure-jwt-secret"
PORT=3001
AI_SERVICE_URL="http://localhost:8000"

2. Run Database Migrations

cd backend
npx prisma generate
npx prisma migrate dev

3. AI Service Configuration

Configure the Python environment and provide your OpenRouter API key.

cd ai-service
python -m venv venv
source venv/bin/activate  # (or `venv\Scripts\activate` on Windows)
pip install -r requirements.txt

Create ai-service/.env:

OPENROUTER_API_KEY="your-openrouter-key"

4. Start Development Servers

You will need to run three separate processes:

Terminal 1 (Backend API):

cd backend
npm install
npm run start:dev

Terminal 2 (AI Semantic Service):

cd ai-service
source venv/bin/activate
uvicorn main:app --reload --port 8000

Terminal 3 (Frontend UI):

cd frontend
npm install
npm run dev

The application will be available at http://localhost:5173.

---

🤝 Contributing & Workflows

1. Grounded Over Generative: Any new AI features must maintain a strict line-of-sight to the source document. If a feature relies on LLMs "guessing" relationships or extracting structured data unreliably, it does not belong in DocLens.
2. Architecture: The NestJS backend acts as the source of truth for Users, Authentication, and access control. The FastAPI service acts purely as a stateless/semantic worker for ingestion and retrieval. Do not cross these boundaries.
3. UI/UX: Maintain the premium, dark-themed aesthetic. Avoid injecting massive frameworks like Tailwind unless strictly necessary for a specific component.

---

📝 License

© Aishwary Vansh 2026. This project is licensed under the MIT License.