Smart Notes — Semantic Search with SBERT & Streamlit

Search your notes by meaning (not just keywords). This app converts notes to Sentence‑BERT embeddings and ranks results with cosine similarity. It ships with a clean Streamlit UI and a tiny CLI.

🌟 Features

• Reactive UI (Streamlit): type to search, results update instantly
• Add / import / export notes (JSON)
• Categories & tags with filters
• Chunk long notes by word count
• Save / load embeddings for faster startup
• CLI for terminal searches
• CPU‑friendly; GPU optional

---

🚀 Quickstart

Clone

git clone https://github.com/RealTimGFM/SmartNoteSystem.git cd SmartNoteSystem

(Recommended) create a virtual env

python -m venv .venv

Windows

.venv\Scripts\activate

Install deps

pip install -r requirements.txt

Run the app

streamlit run app.py

Open: http://localhost:8501

Note: First run downloads the SBERT model (~100 MB). CPU is fine; ignore CUDA/GPU warnings.

---

🧭 Project Structure

SmartNoteSystem/
├─ app.py                     # Streamlit UI
├─ cli.py                     # Terminal search tool
├─ smartnotes/
│  ├─ __init__.py
│  ├─ embeddings.py           # SBERT loading + encoding
│  ├─ search.py               # cosine search, filters
│  ├─ storage.py              # load/save notes JSON
│  └─ chunking.py             # split long notes
├─ data/
│  └─ notes.json              # sample notes (edit/replace)
├─ models/                    # optional saved embeddings
├─ requirements.txt
└─ README.md

---

✍️ Using the App

1. Add notes in the sidebar (or import a JSON).
2. Enter a query (e.g., “How do Python lists work?”).
3. Adjust Top K / Min similarity; filter by Category / Tags.
4. Click Save notes to persist to data/notes.json.
5. Optionally Save embeddings to models/note_embeddings.npy (and Load later).

🖥️ CLI Usage

python cli.py --query "object oriented programming" --top_k 5 --min_similarity 0.4

---

⚙️ How It Works

• Model: all-MiniLM-L6-v2 (Sentence-Transformers) → 384‑dim embeddings
• Similarity: cosine similarity (dot product on L2‑normalized vectors)
• Ranking: sort by similarity, highest first

---

🧩 Tips & Performance

• Keep notes focused on a single concept for best matches.
• Chunk long documents to improve recall.
• Save embeddings to models/ to avoid recomputation after restart.
• Use Category and Tags for quick metadata filtering.

---

❓ Troubleshooting

• Model download fails → check internet; retry; streamlit cache clear
• No results → lower Min similarity, add more notes, or chunk long text
• High RAM → keep notes concise; save & load embeddings
• Windows symlink warnings → safe to ignore

---

📜 License

MIT — do anything