🧭 PaperPilot

AI-powered research paper analysis engine — upload any paper or paste an arXiv ID and get a full scholarly breakdown in seconds.

---

What is PaperPilot?

PaperPilot is a full-stack AI web application that helps students, researchers, and engineers understand research papers faster. Instead of spending hours reading dense academic text, you get a structured, adversarial breakdown of any paper — including what the authors didn't say.

Upload a PDF or paste an arXiv ID. PaperPilot does the rest.

---

Features

Two input modes

• Upload any research paper PDF (up to 35MB)
• Paste an arXiv URL or ID (2301.07041, arxiv.org/abs/2301.07041, or full URL)

Full paper analysis via Gemini 3.5 Flash

• Executive TL;DR (under 150 words)
• In-depth executive summary (3–5 paragraphs)
• Main scholarly contributions
• Methodology breakdown with novelty assessment
• Explicit and implicit assumptions audit
• Weaknesses and structural limitations
• Skeptical peer-reviewer questions
• Future research directions
• Layperson analogy explanation
• Confidence scoring (high / medium / low)

Context-grounded chat

• Ask questions directly about the paper
• Strictly refuses to answer anything not present in the paper text
• No hallucination, no external knowledge injection

Related papers

• Automatically searches arXiv for related work using extracted keywords
• One-click analysis of any related paper

Export

• Download full analysis as a formatted PDF (A4, print-ready)

---

Tech Stack

Layer	Technology
Frontend	React 19, TypeScript, Tailwind CSS, Vite
Backend	Express.js, Node.js
AI	Google Gemini 3.5 Flash (@google/genai)
PDF Parsing	pdfjs-dist (in-memory, no disk writes)
arXiv Integration	arXiv API + fast-xml-parser
PDF Export	html2pdf.js
Deployment	Vercel

---

Getting Started

Prerequisites

• Node.js 18+
• A Gemini API key from Google AI Studio

Local setup

git clone https://github.com/YOUR_USERNAME/paperpilot.git
cd paperpilot
npm install

Create a .env file in the root:

GEMINI_API_KEY=your_key_here

Start the dev server:

npm run dev

Open http://localhost:3000

Deploy to Vercel

1. Push repo to GitHub
2. Import project on vercel.com
3. Set framework to Vite, build command vite build, output dir dist
4. Add GEMINI_API_KEY environment variable
5. Deploy

---

Project Structure

paperpilot/
├── api/                    # Vercel serverless functions
│   ├── paper-upload.ts     # PDF upload + arXiv analysis endpoint
│   ├── arxiv-metadata.ts   # arXiv metadata pre-fetch
│   └── chat.ts             # Context-grounded Q&A endpoint
├── src/
│   ├── components/         # React UI components
│   │   ├── AnalysisTabs.tsx
│   │   ├── DownloadButton.tsx
│   │   ├── InteractiveChat.tsx
│   │   ├── LandingHero.tsx
│   │   ├── RelatedPapers.tsx
│   │   └── SidebarInfo.tsx
│   └── lib/
│       ├── gemini.ts       # Gemini API integration + structured schema
│       ├── arxiv.ts        # arXiv paper fetching
│       └── arxiv-search.ts # Related papers search
├── server.ts               # Express dev server
└── vercel.json             # Vercel routing config

---

How It Works

1. User uploads a PDF or enters an arXiv ID
2. PDF text is extracted in-memory using pdfjs-dist
3. Extracted text is sent to Gemini 3.5 Flash with a structured JSON schema prompt
4. Gemini returns a fully typed analysis object
5. arXiv API is queried for related papers using extracted keywords
6. User can chat with the paper — all answers are strictly grounded in extracted text

---

Sample Analysis

Paper: Verifiable Fully Homomorphic Encryption (arXiv:2301.07041)

TL;DR: This paper establishes the first holistic framework for maliciously-secure Verifiable FHE, mitigating severe key-recovery attacks enabled by FHE's inherent malleability...

Confidence: High

Reviewer Question: How does the proposed delay-and-reduce optimization handle edge cases where the ZKP field size is insufficient to accumulate the required number of RNS splits without overflow?

---

Built By

Ipshita Bhardwaj — 2nd year BTech, Robotics & AI

---

_{Not affiliated with arXiv or Google}