---
title: Substack Semantic Search
emoji: 🔎
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: "6.0.0"
app_file: app.py
pinned: false
---

# 🔎 Semantic Search over Substack Posts

This Space hosts a semantic search engine built over a collection of Substack HTML posts.  
It uses **SentenceTransformers**, **FAISS**, and **Gradio** to provide fast, offline semantic similarity search.

---

## 🚀 How It Works

### 1. **Chunk + Embed**
HTML posts from the `posts/` directory are:
- parsed with BeautifulSoup  
- split into manageable text chunks  
- embedded using `all-MiniLM-L6-v2`  
- stored in a FAISS vector index  

### 2. **Vector Search**
At runtime, the app:
- loads `faiss_index.bin` and `faiss_meta.pkl`
- embeds the user query
- retrieves the most semantically relevant chunks

### 3. **Gradio App**
The search UI is powered by Gradio and runs fully offline inside this Space.

---

## Local Usage

To rebuild the FAISS index locally:

```
pip install -r requirements.txt
python src/build_index.py
python app.py
````
Ensure your `.html` files live in:
```
posts/
```

Make sure these files are at root

```
faiss_index.bin
faiss_meta.pkl
app.py
requirements.txt
```

Once your local run works:
```
python app.py
```