🧠 NoteWeb

A local-first AI-powered tool to search, summarize, and understand your notes — built for students, by a student.
Uses chunking, vector embeddings, and LLaMA 3 via Ollama for real semantic understanding.

---

Features

• Index PDFs, Word, Excel, and PowerPoint files into semantic chunks
• Embed and store those chunks using vector search
• Ask questions using LLaMA 3 via Ollama (offline + local)
• Get answers with real context grounding (like the ChatGPT Retrieval Plugin)

• • Follow-up question support with persistent memory
• • Ingest multiple files at once from a folder
• • Test-ready with test files/ folder

---

Folder Structure

noteweb/
├── main.py                 # CLI entrypoint
├── search.py              # Embedding search logic
├── llm_answerer.py        # Sends chunks to LLaMA via Ollama
├── embedder.py            # Generates embeddings
├── chunker.py             # Breaks files into semantic chunks
├── files_loader.py        # PDF loader (more formats soon)
├── generate_index.py      # Index generator (embeds + saves)
├── embeddings_index.json  # Your saved vector index
├── test files/            # Sample PDFs to test with
├── requirements.txt
└── venv/                  # Your virtual environment

---

Requirements

• Python 3.11+
• Ollama installed and running

  brew install ollama
  ollama run llama3

• Install dependencies:

  pip install -r requirements.txt

---

How to Use

1. Clone the Repository

git clone https://github.com/marcanjoul/noteweb.git
cd noteweb

Or, if you downloaded the ZIP, unzip it and navigate into the folder via:

cd ~/Desktop/noteweb-main  # or wherever you saved it

2. Set Up a Virtual Environment

We recommend using a virtual environment to keep dependencies clean:

python3 -m venv venv
source venv/bin/activate  # (for mac) # On Windows, use venv\Scripts\activate

3. Install Dependencies

Run the following to install all required packages:

pip install -r requirements.txt

If needed, manually install these extras:

pip install python-docx python-pptx openpyxl sentence-transformers

4. Add Your Files

Create or drop any files you want to search into the test files/ directory. Supported formats: .pdf .docx .pptx .xlsx

5. Run this to generate semantic embeddings from files in your folder

python generate_index.py

This will:

• Load all .pdf, .docx, .xslx, and .pptx files from your test files/ folder
• Chunk the content into semantically meaningful parts
• Embed each chunk using sentence-transformers
• Save everything to embeddings_index.json

6. Search Your Files with the AI

python search.py

This will:

• Search your indexed chunks for relevant context

• Pass top matches to LLaMA 3

• Return an answer based on your notes

• You can also ask follow-up questions, as NoteWeb remembers the context!

  💡 Example Usage

  What is the difference between supervised and unsupervised learning?

  ❗Requirements Make sure you have:

  • Python 3.9+
  • pip
  • Optional: Ollama installed and running (for local LLaMA 3 support)

  💡 Optional: Skip venv (Not Recommended) You can also run NoteWeb without using a virtual environment:

  pip install -r requirements.txt
  pip install python-docx python-pptx openpyxl sentence-transformers

7. Ask a question

python main.py --search "What is instruction-level parallelism?"

---

Why This Matters

NoteWeb simulates real retrieval-augmented generation (RAG) — the same strategy used in:

• ChatGPT w/ File Uploads
• Perplexity AI
• Open-source RAG pipelines (like LangChain, LlamaIndex)

But here, it’s all:

• Local
• Educational
• Hackable

Perfect for learning how vector search + LLMs work together.

---

File Support

• ✅ PDF

• • ✅ DOCX (.docx)
• • ✅ PowerPoint (.pptx)
• • ✅ Excel (.xlsx)
• • 🔜 TXT, Markdown, Web scraping

You can drop files into the test files/ folder!

---

Roadmap

• Multi-file indexing (entire folders at once)
• DOCX, PPTX, and XLSX support
• Follow-up question support
• Optional toggle in code for chunk visibility
• Index caching to skip re-embedding unchanged files
• Command-line UI / TUI
• Web UI

---

Project Status

NoteWeb is my first AI-integrated project — made while learning:

• How LLMs like LLaMA work
• What “semantic search” really means
• How chunking, embeddings, and vector stores come together

This is the foundation for bigger projects — search tools, academic companions, even personalized AI.

---

Credits

• Built with 💻 and ☕ by @marcanjoul
• PDF parsing via PyMuPDF
• DOCX parsing via python-docx
• PowerPoint parsing via python-pptx
• Excel parsing via openpyxl
• Embeddings via sentence-transformers
• LLM answers via Ollama and Meta’s LLaMA 3

---

Want to improve or collaborate?

Open an issue, drop a PR, or fork it and make it your own.I'd appreciate any feedback!