PrivateDoc AI

A document-grounded AI assistant built with Next.js 16 and Retrieval-Augmented Generation (RAG).

Live Demo

https://private-doc.vercel.app

Problem

Modern LLM-based tools such as OpenAI's ChatGPT or Google’s Gemini are powerful, but when used for document Q&A they often:

• Hallucinate answers
• Use outside knowledge
• Cannot guarantee document grounding
• Store user data
• Fail to provide source traceability

Solution

PrivateDoc AI solves this by:

• Restricting answers strictly to uploaded document
• Returning a fallback if answer is not present
• Highlighting source text inside the PDF
• Avoiding document storage
• Providing controlled retrieval

System Architecture

flowchart TD
    A[User Uploads PDF] --> B[PDF Parsing]
    B --> C[Text Chunking]
    C --> D[Embedding Generation]
    D --> E[Vector Similarity Search]
    E --> F[Top-K Relevant Chunks]
    F --> G[Strict Prompt Construction]
    G --> H[LLM Response]
    H --> I[Answer + Confidence Score]
    F --> J[Highlight Source in PDF Viewer]

Loading

Architecture Overview

Layer	Technology Used	Purpose
Frontend	Next.js 16 (App Router)	UI + Routing
Styling	Tailwind CSS	Modern SaaS UI
PDF Viewer	PDF.js / react-pdf	Document rendering
Markdown	ReactMarkdown	Response formatting
Backend	Custom RAG Pipeline	Controlled retrieval
Embeddings	Gemini / Groq API	Vector representation
Retrieval	Top-K Similarity Search	Relevant context filtering

RAG Pipeline Design

1. Chunking Strategy

• Fixed-size chunks with overlap
• Preserves semantic boundaries
• Optimized for embedding efficiency

2. Embeddings

• Generated using Gemini / Groq APIs
• Stored temporarily in memory (not persisted)

3. Retrieval

• Cosine similarity search
• Top-k relevant chunks selected
• Strict filtering to avoid noise

4. Prompt Constraint Layer

The LLM receives:

• Only retrieved chunks
• Explicit instruction:
• “Answer strictly from context.”
• “If not found, return: The document does not contain this information.”

Core Features

• Upload any PDF
• Automatic document summary
• 3 context-aware suggested questions
• Custom question support
• Strict grounded answers
• No hallucination policy
• Confidence score
• Highlight exact source text
• Independent scroll UX
• Clean SaaS dashboard UI
• Zero document storage

How It’s Different from ChatGPT

Feature	ChatGPT	PrivateDoc AI
Grounded to document	❌	✅
Hallucination fallback	❌	✅
Source highlighting	❌	✅
Controlled RAG	❌	✅
No document storage	❌	✅

Local Setup

npm install
npm run dev

Environment Variables

Create a .env.local file:

GOOGLE_API_KEY=your_key

Example Fallback Behavior

If a question is asked that is not present in the document:

Response: "The document does not contain this information." This ensures:

• Zero hallucination
• Strict grounding
• Predictable enterprise behavior

Future Improvements

• Persistent vector database (optional secure mode)

• Multi-document querying

• Role-based access control

• Citation export

• Enterprise deployment mode

Author

Rohan Sharma