PaperPal

Instant academic paper formatting — powered by LLMs and rule-based LaTeX generation.

Upload a research paper, pick a citation style, and get a publication-ready LaTeX PDF. Two pipelines: a fast LLM-driven converter for .docx/.txt/.tex files, and a Pro pipeline that parses PDFs with layout-aware extraction and generates LaTeX through pure rule-based code — zero hallucination.

Screenshots

Landing Page

Format Selection

Upload Your Paper

Split-Pane Editor with Live Preview

PDF Download

Why PaperPal?

Every researcher has been there — the paper is done, the content is solid, but reformatting for APA, IEEE, or Vancouver eats hours. Adjusting margins, citation styles, heading levels, and reference lists manually is tedious work that has nothing to do with actual research.

PaperPal automates this entirely. Drop in your document, select a format, and the system produces LaTeX output that follows the exact typographic and structural rules of your chosen style. The generated PDF is ready to submit.

Supported Formats

Format	Full Name	Typical Fields
APA	American Psychological Association (7th ed.)	Psychology, Education, Social Sciences
MLA	Modern Language Association (9th ed.)	Humanities, Literature, Arts
Chicago	Chicago Manual of Style (Notes & Bibliography)	History, Publishing, General Academic
Harvard	Harvard Referencing	Business, Social Sciences, General Use
IEEE	IEEE Conference / Journal	Engineering, Computer Science, Electronics
AMA	American Medical Association	Medicine, Health, Biological Sciences
Vancouver	Vancouver (ICMJE)	Biomedical Journals, Clinical Research
ACS	American Chemical Society	Chemistry, Biochemistry, Materials Science
CSE	Council of Science Editors	Biology, Earth Sciences, Natural Sciences
Custom	User-Defined	Any — define your own rules

Each format has a dedicated master prompt describing the exact visual and structural expectations — fonts, spacing, heading hierarchy, citation mechanics, reference list formatting.

Two Pipelines

Standard Pipeline

Upload .docx, .txt, or .tex files. The LLM converts your content into LaTeX for the selected format via Server-Sent Events with real-time progress tracking.

Pro Pipeline (PDF-to-LaTeX)

Upload a PDF research paper. The system:

Extracts content — text, images, tables, and equations using unpdf + pdfjs-dist
Analyzes structure with AI — sends extracted text to an LLM to identify title, authors, abstract, sections, references, and metadata as structured JSON
Generates LaTeX with rules — passes the structured JSON to one of 9 format-specific rule-based LaTeX generators (pure TypeScript, zero LLM calls, zero hallucination)
Compiles to PDF — sends the LaTeX to TeXLive.net's free API and returns a compiled PDF

The Pro pipeline is completely isolated from the standard pipeline — separate routes, separate pages, separate code.

Architecture

                               ┌───────────────────────────┐
                               │     Next.js Frontend       │
                               │     (React 19 + SSR)       │
                               └─────────────┬─────────────┘
                                             │
                ┌────────────────────────────┼────────────────────────────┐
                │            STANDARD        │           PRO              │
                ▼                            │           ▼                │
    ┌───────────────────┐                    │  ┌─────────────────┐      │
    │  /api/parse        │                    │  │ /api/pro/parse   │      │
    │  Mammoth (.docx)   │                    │  │ unpdf (PDF)      │      │
    │  + .txt / .tex     │                    │  │ Images + Tables  │      │
    └────────┬──────────┘                    │  │ + Equations      │      │
             │                               │  └────────┬────────┘      │
             ▼                               │           ▼               │
    ┌───────────────────┐                    │  ┌─────────────────┐      │
    │  /api/convert      │                    │  │ /api/pro/extract │      │
    │  SSE streaming     │                    │  │ LLM → JSON       │      │
    │  LLM → LaTeX       │                    │  │ (Edge, SSE)      │      │
    │  (Edge runtime)    │                    │  └────────┬────────┘      │
    └────────┬──────────┘                    │           ▼               │
             │                               │  ┌─────────────────┐      │
             │                               │  │ Rule-Based       │      │
             │                               │  │ LaTeX Generators │      │
             │                               │  │ (9 formats, TS)  │      │
             │                               │  └────────┬────────┘      │
             │                               │           ▼               │
             │                               │  ┌─────────────────┐      │
             │                               │  │ /api/pro/compile │      │
             │                               │  │ TeXLive.net API  │      │
             │                               │  │ LaTeX → PDF      │      │
             │                               │  └────────┬────────┘      │
             ▼                               │           ▼               │
    ┌──────────────────────────────────────────────────────────────────┐
    │                    Browser Preview + PDF Download                 │
    └──────────────────────────────────────────────────────────────────┘

    ┌───────────────┐    ┌──────────────────────────────────┐
    │  Token Pool    │    │  HuggingFace Inference API        │
    │  5 HF tokens   │───▶│  Qwen 72B → Llama 70B → Mixtral  │
    │  Round-robin   │    │  → Qwen Coder 32B → Gemma 2B     │
    └───────────────┘    └──────────────────────────────────┘

    ┌───────────────┐    ┌───────────────┐
    │  MongoDB Atlas │    │  JWT + bcrypt  │
    │  User accounts │    │  httpOnly auth │
    └───────────────┘    └───────────────┘

Key Design Decisions

Token round-robin — 5 HuggingFace API tokens rotate to avoid per-token rate limits. Rate-limited (429) or unavailable (503) tokens automatically cycle to the next.
Model fallback chain — Qwen 72B → Llama 3.3 70B → Qwen Coder 32B → Mixtral 8x7B → Gemma 2 2B. Best quality is tried first, with automatic fallback.
Rule-based LaTeX generation (Pro) — Each of the 9 citation formats has a dedicated TypeScript generator (150–340 lines each) encoding the exact formatting rules. No LLM is involved in LaTeX code generation — eliminating hallucinated citations, duplicate content, and formatting errors.
Code-based references — References and bibliography entries are parsed and formatted programmatically in both pipelines.
Anti-hallucination guards — Strict grounding rules are injected into every LLM prompt. The LLM reformats existing content without inventing text, fake authors, or placeholder references.
Edge runtime for streaming — The /api/convert and /api/pro/extract routes use the Vercel Edge runtime, enabling SSE streaming without the 10-second Node.js timeout limit.
JSON repair — The Pro pipeline includes a repairTruncatedJSON() function that can fix common LLM output issues (unclosed strings, unbalanced braces) and a full fallback parser for when JSON parsing fails entirely.

Tech Stack

Layer	Technology
Framework	Next.js 15 (App Router, React 19)
Language	TypeScript 5.7
Styling	Tailwind CSS 4
Animations	Framer Motion
Icons	Lucide React
Document Parsing	Mammoth (DOCX), unpdf (PDF)
PDF Content Extraction	unpdf + pdfjs-dist (text, images, tables, equations)
LLM Inference	HuggingFace Inference API
LaTeX Compilation	TeXLive.net (free, no API key)
Database	MongoDB Atlas via Mongoose
Auth	JWT + bcryptjs
Deployment	Vercel

Getting Started

Prerequisites

Node.js >= 18.x (download)
npm >= 9.x (comes with Node)
HuggingFace API tokens — free tier works (create tokens here)
MongoDB Atlas cluster — free tier works (create one here)

Installation

git clone https://github.com/davesohamm/PaperPal_Hack.git
cd PaperPal_Hack
npm install

Environment Variables

cp .env.example .env.local

Edit .env.local:

# HuggingFace tokens (1-5, more = fewer rate limits)
HF_TOKEN_1=hf_your_first_token
HF_TOKEN_2=hf_your_second_token
HF_TOKEN_3=hf_your_third_token
HF_TOKEN_4=hf_your_fourth_token
HF_TOKEN_5=hf_your_fifth_token

# MongoDB connection string
MONGODB_URI=mongodb+srv://user:password@cluster.mongodb.net/paperpal

# JWT secret (any random string, 32+ characters)
JWT_SECRET=your-secret-key-here

HuggingFace token setup:

Go to huggingface.co/settings/tokens
Create a Fine-grained token
Check "Make calls to Inference Providers"
Copy and paste into .env.local

The system works with as few as 1 token, but rate limits will be hit more often.

MongoDB Atlas setup:

Create a free cluster at mongodb.com/atlas
Add a database user with read/write access
Whitelist your IP (or 0.0.0.0/0 for dev)
Copy the connection string, replace <db_password>, append /paperpal

Running Locally

# On Windows, increase memory for large builds:
$env:NODE_OPTIONS="--max-old-space-size=4096"; npm run dev

# On macOS/Linux:
NODE_OPTIONS="--max-old-space-size=4096" npm run dev

Open http://localhost:3000.

For production build:

npm run build
npm start

How It Works

Standard Pipeline

Pick a format — Choose from 10 citation styles on the formats page
Upload your document — Drag and drop a .docx, .txt, or .tex file
AI conversion — Text is chunked and sent to HuggingFace LLMs via SSE streaming with real-time progress
LaTeX assembly — Preamble is generated first, body chunks are converted, references are formatted by code (not LLM)
Preview and download — Split-pane editor with raw LaTeX on the left, rendered preview on the right

Pro Pipeline (PDF input)

Upload PDF at /pro/upload — select target format, upload your source PDF
PDF parsing — unpdf + pdfjs-dist extracts text per page, detects images (with pixel data), tables (heuristic column alignment), and equations (regex pattern matching)
AI structure extraction — Extracted text is sent to the LLM to produce a compact structured JSON: title, authors, abstract, keywords, sections, references, metadata
Section enrichment — The LLM's compact summaries are enriched with full text from the original extraction by matching section headings back to the source
Rule-based LaTeX generation — The appropriate format generator (e.g., ieee.ts, apa.ts) converts the structured features into complete, compilable LaTeX — no LLM involved
Compilation — LaTeX is sent to TeXLive.net's free API, which returns a compiled PDF
Editor — Split view with editable LaTeX code, PDF preview, download buttons for both .tex and .pdf

Project Structure

PaperPal_Hack/
├── public/
│   └── favicon.svg
├── images/                              # Demo screenshots
├── src/
│   ├── app/
│   │   ├── api/
│   │   │   ├── auth/                    # Authentication routes
│   │   │   │   ├── route.ts             # Unified auth endpoint
│   │   │   │   ├── signin/route.ts      # Sign in
│   │   │   │   ├── signup/route.ts      # Sign up
│   │   │   │   ├── me/route.ts          # Token verification
│   │   │   │   └── logout/route.ts      # Cookie clear
│   │   │   ├── parse/route.ts           # Document parsing (Mammoth)
│   │   │   ├── convert/route.ts         # LLM LaTeX generation (Edge, SSE)
│   │   │   └── pro/                     # Pro pipeline API
│   │   │       ├── parse/route.ts       # PDF extraction (Node runtime)
│   │   │       ├── extract/route.ts     # LLM feature extraction (Edge, SSE)
│   │   │       └── compile/route.ts     # LaTeX → PDF via TeXLive.net (Edge)
│   │   ├── auth/page.tsx                # Sign in / Sign up
│   │   ├── formats/page.tsx             # Format selection grid
│   │   ├── upload/page.tsx              # File upload (standard)
│   │   ├── custom-format/page.tsx       # Custom format builder
│   │   ├── editor/page.tsx              # LaTeX editor + preview (standard)
│   │   ├── pro/
│   │   │   ├── upload/page.tsx          # PDF upload + format selection (Pro)
│   │   │   └── editor/page.tsx          # LaTeX editor + compiled PDF (Pro)
│   │   ├── page.tsx                     # Landing page
│   │   ├── layout.tsx                   # Root layout + AuthProvider
│   │   └── globals.css                  # Global styles
│   ├── components/
│   │   ├── Navbar.tsx                   # Navigation with Pro badge + user menu
│   │   ├── GlowCard.tsx                 # Animated card component
│   │   ├── PageTransition.tsx           # Route transition wrapper
│   │   └── TextReveal.tsx               # Animated text reveal
│   ├── context/
│   │   └── AuthContext.tsx              # Auth state + JWT management
│   └── lib/
│       ├── constants.ts                 # Format definitions + UI config
│       ├── db.ts                        # MongoDB connection (lazy init)
│       ├── jwt.ts                       # JWT sign/verify helpers
│       ├── models.ts                    # LLM model configs + format prompts
│       ├── token-pool.ts               # HF token rotation + fallback
│       ├── user.ts                      # User schema + bcrypt
│       └── pro/                         # Pro pipeline library
│           ├── types.ts                 # Shared interfaces
│           ├── pdf-extractor.ts         # PDF → text/images/tables/equations
│           ├── llm-prompts.ts           # Structured extraction prompts
│           └── latex-generators/        # Rule-based LaTeX generators
│               ├── base.ts             # Shared utilities (escapeTeX, tables, figures, equations)
│               ├── index.ts            # Generator registry
│               ├── apa.ts             # APA 7th Edition
│               ├── mla.ts             # MLA 9th Edition
│               ├── ieee.ts            # IEEE Conference
│               ├── chicago.ts         # Chicago Manual of Style
│               ├── harvard.ts         # Harvard Referencing
│               ├── ama.ts             # AMA
│               ├── vancouver.ts       # Vancouver
│               ├── acs.ts             # ACS
│               └── cse.ts            # CSE
├── .env.example                        # Environment variable template
├── .gitignore
├── next.config.ts
├── package.json
├── tsconfig.json
└── README.md

Deployment

Vercel (Recommended)

Push to GitHub
Import the repository on vercel.com
Add environment variables in Project Settings → Environment Variables:
- HF_TOKEN_1 through HF_TOKEN_5
- MONGODB_URI
- JWT_SECRET
Deploy — Vercel handles the build automatically

The Edge runtime routes (/api/convert, /api/pro/extract, /api/pro/compile) bypass Vercel's 10-second Node.js timeout, enabling long-running SSE streams.

Navigation Menu

Repository files navigation

PaperPal

Screenshots

Landing Page

Format Selection

Upload Your Paper

Split-Pane Editor with Live Preview

PDF Download

Table of Contents

Why PaperPal?

Supported Formats

Two Pipelines

Standard Pipeline

Pro Pipeline (PDF-to-LaTeX)

Architecture

Key Design Decisions

Tech Stack

Getting Started

Prerequisites

Installation

Environment Variables

Running Locally

How It Works

Standard Pipeline

Pro Pipeline (PDF input)

Project Structure

Deployment

Vercel (Recommended)

Roadmap

Team

License

About

Releases

Packages

Contributors

Languages

Name		Name	Last commit message	Last commit date
images		images
public		public
src		src
.env.example		.env.example
.gitignore		.gitignore
LLM_Rules_for_PapperPal.docx		LLM_Rules_for_PapperPal.docx
NeuralNexus.pptx		NeuralNexus.pptx
README.md		README.md
next.config.ts		next.config.ts
package-lock.json		package-lock.json
package.json		package.json
postcss.config.mjs		postcss.config.mjs
rules_formats.txt		rules_formats.txt
tsconfig.json		tsconfig.json

Latest commit

History

Folders and files

Repository files navigation

PaperPal

Screenshots

Landing Page

Format Selection

Upload Your Paper

Split-Pane Editor with Live Preview

PDF Download

Table of Contents

Why PaperPal?

Supported Formats

Two Pipelines

Standard Pipeline

Pro Pipeline (PDF-to-LaTeX)

Architecture

Key Design Decisions

Tech Stack

Getting Started

Prerequisites

Installation

Environment Variables

Running Locally

How It Works

Standard Pipeline

Pro Pipeline (PDF input)

Project Structure

Deployment

Vercel (Recommended)

Roadmap

Team

License

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages