ScratchLM

Zero-DB AI Study Assistant — Choose your model, upload PDFs, and get AI-powered cheatsheets and interactive tutoring.

✨ Key Features

🤖 Dual Model Support

• gemma-4-31b-it via Google AI Studio — fast, text-based
• nvidia/nemotron-3-nano-omni-30b-a3b-reasoning via NVIDIA NIM — multimodal, PDF rendered as images
• Automatic fallback: if Nemotron fails, gracefully switches to Gemma

📄 PDF Processing

• Drag & Drop Upload: Upload PDFs directly in your browser
• Gemma path: PDF → pypdf text extraction → text sent to model
• Nemotron path: PDF → PyMuPDF renders each page as PNG image → sent to model as image content blocks
• Multi-Page Support: All pages rendered and sent for full document understanding

📚 AI-Powered Cheatsheet Generator

• 4-Zone Structure: Automatically generates comprehensive study guides:
  • Zone A: Concept Deep-Dive
  • Zone B: Real-World Analogies
  • Zone C: Worked Problems
  • Zone D: Exam Traps & Strategy
• Markdown Export: Download cheatsheets as .md files for Obsidian, Notion, or any app

💬 AI Chat with Full Context

• Document-Aware: Chat remembers everything from your uploaded PDFs
• Conversation History: Multi-turn dialogues with context retention
• Step-by-Step Teaching: AI tutor explains concepts rather than just answering

🔒 Privacy First

• Zero Database: No user data stored anywhere
• Session Isolation: Each browser tab is isolated
• One-Click Wipe: Refresh the page to clear all data

---

🎓 AI Architecture

Model Selection

The app ships with two model options, selectable via the UI:

Model	Provider	Document Processing
gemma-4-31b-it	Google AI Studio	PDF → pypdf text extraction → text
nvidia/nemotron-3-nano-omni-30b-a3b-reasoning	NVIDIA NIM	PDF → PNG images (150 DPI) → multimodal

Retry & Fallback

Nemotron calls retry up to 2 times with a 2-second delay on failure. If all retries fail, the request automatically falls back to Gemma — no data loss.

---

🛠️ Technical Stack

Component	Technology
Web Framework	FastAPI
Frontend	Vanilla HTML/CSS/JS
AI Models	Gemma 4 31B IT (Google AI Studio) + Nemotron 3 Nano (NVIDIA NIM)
SDK	OpenAI Python SDK (both endpoints are OpenAI-compatible)
PDF Text Extraction	pypdf
PDF Image Rendering	PyMuPDF (fitz)
Markdown Rendering	marked.js

---

🚀 Getting Started

Prerequisites

• A Google AI Studio API key (GEMINI_API_KEY) — free tier available
• A NVIDIA NIM API key (NVIDIA_API_KEY) — available at build.nvidia.com
• Python 3.12+ or Docker

Local Setup

1. Clone the repository:

   git clone https://github.com/mingnatthakitt/ScratchLM.git
   cd ScratchLM

2. Set up environment variables:

   export GEMINI_API_KEY="your_google_api_key_here"
   export NVIDIA_API_KEY="your_nvidia_api_key_here"

3. Run with Docker:

   docker build -t scratchlm .
   docker run -p 7860:7860 \
     -e GEMINI_API_KEY=$GEMINI_API_KEY \
     -e NVIDIA_API_KEY=$NVIDIA_API_KEY \
     scratchlm

   Or run without Docker:

   pip install -r requirements.txt
   python app.py

4. Open in browser: Navigate to http://localhost:7860

Hugging Face Spaces

1. Fork the Space: Clone it to your account
2. Add Secrets: Go to Space Settings → Repository secrets → Add GEMINI_API_KEY and NVIDIA_API_KEY
3. Done: The Space will automatically build and run

---

📖 How to Use

Step 1: Select a Model

Use the dropdown in the sidebar to choose:

• Gemma 4 — text-based extraction, good for fast responses
• Nemotron 3 — renders PDF pages as images for full document understanding

Step 2: Upload PDF

Drag and drop a PDF onto the upload area, or click to browse.

Step 3: Extract

Click ⚡ Extract to process the document. For Gemma, text is extracted. For Nemotron, pages are rendered as images.

Step 4: Generate Cheatsheet or Chat

• ✨ Generate: Create a 4-zone study guide from your PDF
• 💬 Chat: Ask questions about your PDF with full context awareness
• 📥 Download .md: Save the cheatsheet as a Markdown file

---

⚖️ License

GNU Affero General Public License v3.0 (AGPL-3.0)

This program is free software: you can redistribute it and/or modify it under the terms of the GNU Affero General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Affero General Public License for more details.

You should have received a copy of the GNU Affero General Public License along with this program. If not, see https://www.gnu.org/licenses/agpl-3.0.en.html.

---

🙏 Acknowledgments

• Google AI for Gemma models via AI Studio
• NVIDIA for Nemotron models via NIM
• Hugging Face for Spaces infrastructure
• FastAPI for the excellent web framework